Recent studies on LLM agent scaling have highlighted the potential of Multi-Agent Debate (MAD) to enhance reasoning abilities. However, the critical aspect of role allocation strategies remains underexplored. In this study, we demonstrate that allocating roles with differing viewpoints to specific positions significantly impacts MAD's performance in reasoning tasks. Specifically, we find a novel role allocation strategy, "Truth Last", which can improve MAD performance by up to 22% in reasoning tasks. To address the issue of unknown truth in practical applications, we propose the Multi-Agent Debate Consistency (MADC) strategy, which systematically simulates and optimizes its core mechanisms. MADC incorporates path consistency to assess agreement among independent roles, simulating the role with the highest consistency score as the truth. We validated MADC across a range of LLMs (9 models), including the DeepSeek-R1 Distilled Models, on challenging reasoning tasks. MADC consistently demonstrated advanced performance, effectively overcoming MAD's performance bottlenecks and providing a crucial pathway for further improvements in LLM agent scaling.

Autonomous Feature Engineering (AFE) is critical for improving predictive performance on tabular data by relieving humans from manual feature crafting. However, traditional AFE lacks the semantic guidance needed to fully exploit domain knowledge. Although large language models (LLMs) can, in principle, emulate experts, existing approaches typically operate in an open code space that directly generates and rewrites entire features; without a compositional structural representation and invariant constraints, edits are coarse and non-local, making it hard to distill interpretable features with high information content and rich hierarchical structure.

Rong Cheng, Jinyi Liu, Yan Zheng, Fei Ni, Jiazhen Du, Hangyu Mao, Fuzheng Zhang, Bo Wang, Jianye Hao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Reward functions are crucial for policy learning. Large Language Models (LLMs), with strong coding capabilities and valuable domain knowledge, provide an automated solution for high-quality reward design. However, code-based reward functions require precise guiding logic and parameter configurations within a vast design space, leading to low optimization efficiency. To address the challenges, we propose an efficient automated reward design framework, called R, which decomposes reward design into two parts: reward structure evolution and parameter alignment optimization. To design high-quality reward structures, R maintains a reward function population and modularizes the functional components. LLMs are employed as the mutation operator, and module-level crossover is proposed to facilitate efficient exploration and exploitation. To design more efficient reward parameters, R first leverages LLMs to generate multiple critic functions for trajectory comparison and annotation. Based on these critics, a voting mechanism is employed to collect the trajectory segments with high-confidence labels. These labeled segments are then used to refine the reward function parameters through preference learning. Experiments on diverse robotic control tasks demonstrate that R outperforms strong baselines in both reward design efficiency and quality, surpassing human-designed reward functions.

The integration of evolutionary algorithms (EAs) with reinforcement learning (RL) has shown superior performance compared to standalone methods. However, previous research focuses on exploration in policy parameter space, while overlooking the reward function search. To bridge this gap, we propose **LaRes**, a novel hybrid framework that achieves efficient policy learning through reward function search. LaRes leverages large language models (LLMs) to generate the reward function population, guiding RL in policy learning. The reward functions are evaluated by the policy performance and improved through LLMs. To improve sample efficiency, LaRes employs a shared experience buffer that collects experiences from all policies, with each experience containing rewards from all reward functions. Upon reward function updates, the rewards of experiences are relabeled, enabling efficient use of historical data. Furthermore, we introduce a Thompson sampling-based selection mechanism that enables more efficient elite interaction. To prevent policy collapse when improving reward functions, we propose the reward scaling and parameter constraint mechanisms to efficiently coordinate reward search with policy learning. Across both initialized and non-initialized settings, LaRes consistently achieves state-of-the-art performance, outperforming strong baselines in both sample efficiency and final performance. The code is available at https://github.com/yeshenpy/LaRes.

Many real-world control problems require continual policy adjustments to balance multiple objectives, which requires the acquisition of high-quality policies to cover diverse preferences. Multi-Objective Reinforcement Learning (MORL) provides a general framework to solve such problems. However, current MORL methods suffer from high sample complexity, primarily due to the neglect of efficient knowledge sharing and conflicts in optimization with different preferences. To this end, this paper introduces a novel framework, Conflict Objective Regularization in Latent Space (**COLA**). To enable efficient knowledge sharing, COLA establishes a shared latent representation space for common knowledge, which can avoid redundant learning under different preferences. Besides, COLA introduces a regularization term for the value function to mitigate the negative effects of conflicting preferences on the value function approximation, thereby improving the accuracy of value estimation. The experimental results across various multi-objective continuous control tasks demonstrate the significant superiority of COLA over the state-of-the-art MORL baselines. Code is available at https://github.com/yeshenpy/COLA.

Spreadsheets are ubiquitous across the World Wide Web, playing a critical role in enhancing work efficiency across various domains. Large language model (LLM) has been recently attempted for automatic spreadsheet manipulation but has not yet been investigated in complicated and realistic tasks where reasoning challenges exist (e.g., long horizon manipulation with multi-step reasoning and ambiguous requirements). To bridge the gap with the real-world requirements, we introduce SheetRM, a benchmark featuring long-horizon and multi-category tasks with reasoning-dependent manipulation caused by real-life challenges. To mitigate the above challenges, we further propose SheetAgent, a novel autonomous agent that utilizes the power of LLMs. SheetAgent consists of three collaborative modules: Planner, Informer, and Retriever, achieving both advanced reasoning and accurate manipulation over spreadsheets without human interaction through iterative task reasoning and reflection. Extensive experiments demonstrate that SheetAgent delivers 20--40% pass rate improvements on multiple benchmarks over baselines, achieving enhanced precision in spreadsheet manipulation and demonstrating superior table reasoning abilities. More details and visualizations are available at the https://sheetagent.github.io/. The datasets and source code are available at https://anonymous.4open.science/r/SheetAgent.

Evolutionary reinforcement learning (ERL), which integrates the evolutionary algorithms (EAs) and reinforcement learning (RL) for optimization, has demonstrated remarkable performance advancements. By fusing both the approaches, ERL has emerged as a promising research direction. This survey offers a comprehensive overview of the diverse research branches in ERL. Specifically, we systematically summarize the recent advancements in related algorithms and identify three primary research directions: 1) EA-assisted optimization of RL; 2) RL-assisted optimization of EA; and 3) synergistic optimization of EA and RL. Following that, we conduct an in-depth analysis of each research direction, organizing multiple research branches. We elucidate the problems that each branch aims to tackle and how the integration of EAs and RL addresses these challenges. In conclusion, we discuss potential challenges and prospective future research directions across various research directions. To facilitate researchers in delving into ERL, we organize the algorithms and codes involved onhttps://github.com/yeshenpy/Awesome-Evolutionary-Reinforcement-Learning.

In reinforcement learning, the optimism in the face of uncertainty (OFU) is a mainstream principle for directing exploration towards less explored areas, characterized by higher uncertainty. However, in the presence of environmental stochasticity (noise), purely optimistic exploration may lead to excessive probing of high-noise areas, consequently impeding exploration efficiency. Hence, in exploring noisy environments, while optimism-driven exploration serves as a foundation, prudent attention to alleviating unnecessary over-exploration in high-noise areas becomes beneficial. In this work, we propose Optimistic Value Distribution Explorer (OVD-Explorer) to achieve a noise-aware optimistic exploration for continuous control. OVD-Explorer proposes a new measurement of the policy's exploration ability considering noise in optimistic perspectives, and leverages gradient ascent to drive exploration. Practically, OVD-Explorer can be easily integrated with continuous control RL algorithms. Extensive evaluations on the MuJoCo and GridChaos tasks demonstrate the superiority of OVD-Explorer in achieving noise-aware optimistic exploration.

AlphaZero and MuZero have achieved state-of-the-art (SOTA) performance in a wide range of domains, including board games and robotics, with discrete and continuous action spaces. However, to obtain an improved policy, they often require an excessively large number of simulations, especially for domains with large action spaces. As the simulation budget decreases, their performance drops significantly. In addition, many important real-world applications have combinatorial (or exponential) action spaces, making it infeasible to search directly over all possible actions. In this paper, we extend AlphaZero and MuZero to learn and plan in more complex multiagent (MA) Markov decision processes, where the action spaces increase exponentially with the number of agents. Our new algorithms, MA Gumbel AlphaZero and MA Gumbel MuZero, respectively without and with model learning, achieve superior performance on cooperative multiagent control problems, while reducing the number of environmental interactions by up to an order of magnitude compared to model-free approaches. In particular, we significantly improve prior performance when planning with much fewer simulation budgets. The code and appendix are available at https://github.com/tjuHaoXiaotian/MA-MuZero.

Despite many breakthroughs in recent years, it is still hard for MultiAgent Reinforcement Learning (MARL) algorithms to directly solve complex tasks in MultiAgent Systems (MASs) from scratch. In this work, we study how to use Automatic Curriculum Learning (ACL) to reduce the number of environmental interactions required to learn a good policy. In order to solve a difficult task, ACL methods automatically select a sequence of tasks (i.e., curricula). The idea is to obtain maximum learning progress towards the final task by continuously learning on tasks that match the current capabilities of the learners. The key question is how to measure the learning progress of the learner for better curriculum selection. We propose a novel ACL framework, PrOgRessive mulTiagent Automatic curricuLum (PORTAL), for MASs. PORTAL selects curricula according to two critera: 1) How difficult is a task, relative to the learners’ current abilities? 2) How similar is a task, relative to the final task? By learning a shared feature space between tasks, PORTAL is able to characterize different tasks based on the distribution of features and select those that are similar to the final task. Also, the shared feature space can effectively facilitate the policy transfer between curricula. Experimental results show that PORTAL can train agents to master extremely hard cooperative tasks, which can not be achieved with previous state-of-the-art MARL algorithms.

Transfer learning (TL) has shown great potential to improve Reinforcement Learning (RL) efficiency by leveraging prior knowledge in new tasks. However, much of the existing TL research focuses on transferring knowledge between tasks that share the same state-action spaces. Further, transfer from multiple source tasks that have different state-action spaces is more challenging and needs to be solved urgently to improve the generalization and practicality of the method in real-world scenarios. This paper proposes TURRET (Transfer Using gRaph neuRal nETworks), to utilize the generalization capabilities of Graph Neural Networks (GNNs) to facilitate efficient and effective multi-source policy transfer learning in the state-action mismatch setting. TURRET learns a semantic representation by accounting for the intrinsic property of the agent through GNNs, which leads to a unified state embedding space for all tasks. As a result, TURRET achieves more efficient transfer with strong generalization ability between different tasks and can be easily combined with existing Deep RL algorithms. Experimental results show that TURRET significantly outperforms other TL methods on multiple continuous action control tasks, successfully transferring across robots with different state-action spaces.

Robotics agents often struggle to understand and follow the multi-modal prompts in complex manipulation scenes which are challenging to be sufficiently and accurately described by text alone. Moreover, for long-horizon manipulation tasks, the deviation from general instruction tends to accumulate if lack of intermediate guidance from high-level subgoals. For this, we consider can we generate subgoal images before act to enhance the instruction following in long-horizon manipulation with multi-modal prompts? Inspired by the great success of diffusion model in image generation tasks, we propose a novel hierarchical framework named as CoTDiffusion that incorporates diffusion model as a high-level planner to convert the general and multimodal prompts into coherent visual subgoal plans, which further guide the low-level policy model before action execution. We design a semantic alignment module that can anchor the progress of generated keyframes along a coherent generation chain, unlocking the chain-of-thought reasoning ability of diffusion model. Additionally, we propose bi-directional generation and frame concat mechanism to further enhance the fidelity of generated subgoal images and the accuracy of instruction following. The experiments cover various robotics manipulation scenarios including visual reasoning, visual rearrange, and visual constraints. CoTDiffusion achieves outstanding performance gain compared to the baselines without explicit subgoal generation, which proves that a subgoal image is worth a thousand words of instruction. The details and visualizations are available at https://cotdiffusion.github.io.

Aligning agent behaviors with diverse human preferences remains a challenging problem in reinforcement learning (RL), owing to the inherent abstractness and mutability of human preferences. To address these issues, we propose AlignDiff, a novel framework that leverages RLHF to quantify human preferences, covering abstractness, and utilizes them to guide diffusion planning for zero-shot behavior customizing, covering mutability. AlignDiff can accurately match user-customized behaviors and efficiently switch from one to another. To build the framework, we first establish the multi-perspective human feedback datasets, which contain comparisons for the attributes of diverse behaviors, and then train an attribute strength model to predict quantified relative strengths. After relabeling behavioral datasets with relative strengths, we proceed to train an attribute-conditioned diffusion model, which serves as a planner with the attribute strength model as a director for preference aligning at the inference phase. We evaluate AlignDiff on various locomotion tasks and demonstrate its superior performance on preference matching, switching, and covering compared to other baselines. Its capability of completing unseen downstream tasks under human instructions also showcases the promising potential for human-AI collaboration. More visualization videos are released on https://aligndiff.github.io/.

Reinforcement Learning with Human Feedback (RLHF) has received significant attention for performing tasks without the need for costly manual reward design by aligning human preferences. It is crucial to consider diverse human feedback types and various learning methods in different environments. However, quantifying progress in RLHF with diverse feedback is challenging due to the lack of standardized annotation platforms and widely used unified benchmarks. To bridge this gap, we introduce **Uni-RLHF**, a comprehensive system implementation tailored for RLHF. It aims to provide a complete workflow from *real human feedback*, fostering progress in the development of practical problems. Uni-RLHF contains three packages: 1) a universal multi-feedback annotation platform, 2) large-scale crowdsourced feedback datasets, and 3) modular offline RLHF baseline implementations. Uni-RLHF develops a user-friendly annotation interface tailored to various feedback types, compatible with a wide range of mainstream RL environments. We then establish a systematic pipeline of crowdsourced annotations, resulting in large-scale annotated datasets comprising more than 15 million steps across 30 popular tasks. Through extensive experiments, the results in the collected datasets demonstrate competitive performance compared to those from well-designed manual rewards. We evaluate various design choices and offer insights into their strengths and potential areas of improvement. We wish to build valuable open-source platforms, datasets, and baselines to facilitate the development of more robust and reliable RLHF solutions based on realistic human feedback. The website is available at https://uni-rlhf.github.io/.

Robotic manipulation tasks often span over long horizons and encapsulate multiple subtasks with different skills. Learning policies directly from long-horizon demonstrations is challenging without intermediate keyframes guidance and corresponding skill annotations. Existing approaches for keyframe identification often struggle to offer reliable decomposition for low accuracy and fail to provide semantic relevance between keyframes and skills. For this, we propose a unified Keyframe Identifier and Skill Anotator (KISA) that utilizes pretrained visual-language representations for precise and interpretable decomposition of unlabeled demonstrations. Specifically, we develop a simple yet effective temporal enhancement module that enriches frame-level representations with expanded receptive fields to capture semantic dynamics at the video level. We further propose coarse contrastive learning and fine-grained monotonic encouragement to enhance the alignment between visual representations from keyframes and language representations from skills. The experimental results across three benchmarks demonstrate that KISA outperforms competitive baselines in terms of accuracy and interpretability of keyframe identification. Moreover, KISA exhibits robust generalization capabilities and the flexibility to incorporate various pretrained representations.

Both Evolutionary Algorithms (EAs) and Reinforcement Learning (RL) have demonstrated powerful capabilities in policy search with different principles. A promising direction is to combine the respective strengths of both for efficient policy optimization. To this end, many works have proposed various mechanisms to integrate EAs and RL. However, it is still unclear which of these mechanisms are complementary and can be fully combined. In this paper, we revisit different mechanisms from five perspectives: 1) Interaction Mode, 2) Individual Architecture, 3) EAs and operators, 4) Impact of EA on RL, and 5) Fitness Surrogate and Usage. We evaluate the effectiveness of each mechanism and experimentally analyze the reasons for the more effective mechanisms. Using the most effective mechanisms, we develop EvoRainbow and EvoRainbow-Exp, which outperform strong baselines and provide state-of-the-art performance across various tasks with distinct characteristics. To promote community development, we release the code on https://github.com/yeshenpy/EvoRainbow.

Reasoning is a fundamental ability towards artificial general intelligence (AGI). Fueled by the success of deep learning, the neural logic machines models (NLMs) have introduced novel neural-symbolic structures and demonstrate great performance and generalization on reasoning and decision-making tasks. However, the original training approaches of the NLMs are still far from perfect, the models would repeat similar mistakes during the training process which leads to sub-optimal performance. To mitigate this issue, we present a novel framework named Failure Reflection Guided Regularizer (FRGR). FRGR first dynamically identifies and summarizes the root cause if the model repeats similar mistakes during training. Then it penalizes the model if it makes similar mistakes in future training iterations. In this way, the model is expected to avoid repeating errors of similar root causes and converge faster to a better-performed optimum. Experimental results on multiple relational reasoning and decision-making tasks demonstrate the effectiveness of FRGR in improving performance, generalization, training efficiency, and data efficiency.

Combining Evolutionary Algorithms (EAs) and Reinforcement Learning (RL) for policy search has been proven to improve RL performance. However, previous works largely overlook value-based RL in favor of merging EAs with policy-based RL. This paper introduces Value-Evolutionary-Based Reinforcement Learning (VEB-RL) that focuses on the integration of EAs with value-based RL. The framework maintains a population of value functions instead of policies and leverages negative Temporal Difference error as the fitness metric for evolution. The metric is more sample-efficient for population evaluation than cumulative rewards and is closely associated with the accuracy of the value function approximation. Additionally, VEB-RL enables elites of the population to interact with the environment to offer high-quality samples for RL optimization, whereas the RL value function participates in the population's evolution in each generation. Experiments on MinAtar and Atari demonstrate the superiority of VEB-RL in significantly improving DQN, Rainbow, and SPR. Our code is available on https://github.com/yeshenpy/VEB-RL.

One of the fundamental challenges for offline reinforcement learning (RL) is ensuring robustness to data distribution. Whether the data originates from a near-optimal policy or not, we anticipate that an algorithm should demonstrate its ability to learn an effective control policy that seamlessly aligns with the inherent distribution of offline data. Unfortunately, behavior regularization, a simple yet effective offline RL algorithm, tends to struggle in this regard. In this paper, we propose a new algorithm that substantially enhances behavior-regularization based on conservative policy iteration. Our key observation is that by iteratively refining the reference policy used for behavior regularization, conservative policy update guarantees gradually improvement, while also implicitly avoiding querying out-of-sample actions to prevent catastrophic learning failures. We prove that in the tabular setting this algorithm is capable of learning the optimal policy covered by the offline dataset, commonly referred to as the in-sample optimal policy. We then explore several implementation details of the algorithm when function approximations are applied. The resulting algorithm is easy to implement, requiring only a few lines of code modification to existing methods. Experimental results on the D4RL benchmark indicate that our method outperforms previous state-of-the-art baselines in most tasks, clearly demonstrate its superiority over behavior regularization.

Diffusion planning has been recognized as an effective decision-making paradigm in various domains. The capability of generating high-quality long-horizon trajectories makes it a promising research direction. However, existing diffusion planning methods suffer from low decision-making frequencies due to the expensive iterative sampling cost. To alleviate this, we introduce DiffuserLite, a super fast and lightweight diffusion planning framework, which employs a planning refinement process (PRP) to generate coarse-to-fine-grained trajectories, significantly reducing the modeling of redundant information and leading to notable increases in decision-making frequency. Our experimental results demonstrate that DiffuserLite achieves a decision-making frequency of $122.2$Hz ($112.7$x faster than predominant frameworks) and reaches state-of-the-art performance on D4RL, Robomimic, and FinRL benchmarks. In addition, DiffuserLite can also serve as a flexible plugin to increase the decision-making frequency of other diffusion planning algorithms, providing a structural design reference for future works. More details and visualizations are available at https://diffuserlite.github.io/.

Leveraging the powerful generative capability of diffusion models (DMs) to build decision-making agents has achieved extensive success. However, there is still a demand for an easy-to-use and modularized open-source library that offers customized and efficient development for DM-based decision-making algorithms. In this work, we introduce CleanDiffuser, the first DM library specifically designed for decision-making algorithms. By revisiting the roles of DMs in the decision-making domain, we identify a set of essential sub-modules that constitute the core of CleanDiffuser, allowing for the implementation of various DM algorithms with simple and flexible building blocks. To demonstrate the reliability and flexibility of CleanDiffuser, we conduct comprehensive evaluations of various DM algorithms implemented with CleanDiffuser across an extensive range of tasks. The analytical experiments provide a wealth of valuable design choices and insights, reveal opportunities and challenges, and lay a solid groundwork for future research. CleanDiffuser will provide long-term support to the decision-making community, enhancing reproducibility and fostering the development of more robust solutions.

Intermittent control problems are common in real world. The interactions between the decision maker and the executor can be discontinuous (intermittent) due to various types of interruptions, e.g. unstable communication channel. Due to intermittent interaction, agents are unable to acquire the state sent by the executor and cannot transmit actions to the executor within a period of time step, i.e. bidirectional blockage, which may lead to inefficiencies of reinforcement learning policies and prevent the executors from completing the task. Such problem is not well studied in the RL community. In this paper, we model Intermittent control problem as an Intermittent Control Markov Decision Process, i.e agents are expected to generate action sequences corresponding to the unavailable states and transmit them before disabling interactions to ensure the smooth and effective motion of executors. However, directly generating multiple future actions in the original action space has unnatural motion issue and exploration difficulty. We propose **M**ulti-step **A**ction **R**epre**S**entation (**MARS**), which encodes a sequence of actions from the original action space to a compact and decodable latent space. Then based on the latent action sequence representation, the mainstream RL methods can be easily optimized to learn a smooth and efficient motion policy. Extensive experiments on simulation tasks and real-world robotic grasping tasks show that MARS significantly improves the learning efficiency and final performances compared with existing baselines.

Long-horizon manipulation tasks with general instructions often implicitly encapsulate multiple sub-tasks, posing significant challenges in instruction following. While language planning is a common approach to decompose general instructions into stepwise sub-instructions, text-only guidance may lack expressiveness and lead to potential ambiguity. Considering that humans often imagine and visualize sub-instructions reasoning out before acting, the imagined subgoal images can provide more intuitive guidance and enhance the reliability of decomposition. Inspired by this, we propose **PERIA**(**PE**rceive, **R**eason, **I**magine, **A**ct), a novel framework that integrates holistic language planning and vision planning for long-horizon manipulation tasks with complex instructions, leveraging both logical and intuitive aspects of task decomposition. Specifically, we first perform a lightweight multimodal alignment on the encoding side to empower the MLLM to perceive visual details and language instructions. The MLLM is then jointly instruction-tuned with a pretrained image-editing model to unlock capabilities of simultaneous reasoning of language instructions and generation of imagined subgoals. Furthermore, we introduce a consistency alignment loss to encourage coherent subgoal images and align with their corresponding instructions, mitigating potential hallucinations and semantic conflicts between the two planning manners. Comprehensive evaluations across three task domains demonstrate that PERIA, benefiting from holistic language and vision planning, significantly outperforms competitive baselines in both instruction following accuracy and task success rate on complex manipulation tasks.

Hybrid action control tasks are common in the real world, which require controlling some discrete and continuous actions simultaneously. To solve these tasks, existing Deep Reinforcement learning (DRL) methods either directly build a separate policy for each type of action or simplify the hybrid action space into a discrete or continuous action control problem. However, these methods neglect the challenge of exploration resulting from the complexity of the hybrid action space. Thus, it is necessary to design more sample efficient algorithms. To this end, we propose a novel Hybrid Control Transformer (Hybrid CtrlFormer), to achieve better exploration and exploitation for the hybrid action control problems. The core idea is: 1) we construct a hybrid action space tree with the discrete actions at the higher level and the continuous parameter space at the lower level. Each parameter space is split into multiple subregions. 2) To simplify the exploration space, a Transformer-based Monte-Carlo tree search method is designed to efficiently evaluate and partition the hybrid action space into good and bad subregions along the tree. Our method achieves state-of-the-art performance and sample efficiency in a variety of environments with discrete-continuous action space.

In this paper, we focus on the decision-making challenge in autonomous driving, a central and intricate problem influencing the safety and practicality of autonomous vehicles. We propose an innovative hierarchical imitation learning framework that effectively alleviates the complexity of learning in autonomous driving decision-making problems by decoupling decision-making tasks into sub-problems. Specifically, the decision-making process is divided into two levels of sub-problems: the upper level directs the vehicle's lane selection and qualitative speed management, while the lower level implements precise control of the driving speed and direction. We harness Transformer-based models for solving each sub-problem, enabling overall hierarchical framework to comprehend and navigate diverse and various road conditions, ultimately resulting in improved decision-making. Through an evaluation in several typical driving scenarios within the SMARTS autonomous driving simulation environment, our proposed hierarchical decision-making framework significantly outperforms end-to-end reinforcement learning algorithms and behavior cloning algorithm, achieving an average pass rate of over 90%. Our framework's effectiveness is substantiated by its commendable achievements at the NeurIPS 2022 Driving SMARTS competition, where it secures dual track championships.

Evolutionary reinforcement learning (ERL) algorithms recently raise attention in tackling complex reinforcement learning (RL) problems due to high parallelism, while they are prone to insufficient exploration or model collapse without carefully tuning hyperparameters (aka meta-parameters). In the paper, we propose a general meta ERL framework via bilevel optimization (BiERL) to jointly update hyperparameters in parallel to training the ERL model within a single agent, which relieves the need for prior domain knowledge or costly optimization procedure before model deployment. We design an elegant meta-level architecture that embeds the inner-level’s evolving experience into an informative population representation and introduce a simple and feasible evaluation of the meta-level fitness function to facilitate learning efficiency. We perform extensive experiments in MuJoCo and Box2D tasks to verify that as a general framework, BiERL outperforms various baselines and consistently improves the learning performance for a diversity of ERL algorithms.

Multi-Agent Reinforcement Learning (MARL) has demonstrated its effectiveness in learning collaboration, but it often struggles with low-quality reward signals and high non-stationarity. In contrast, Evolutionary Algorithm (EA) has shown better convergence, robustness, and signal quality insensitivity. This paper introduces a hybrid framework, Representation Asymmetry and Collaboration Evolution (RACE), which combines EA and MARL for efficient collaboration. RACE maintains a MARL team and a population of EA teams. To enable efficient knowledge sharing and policy exploration, RACE decomposes the policies of different teams controlling the same agent into a shared nonlinear observation representation encoder and individual linear policy representations. To address the partial observation issue, we introduce Value-Aware Mutual Information Maximization to enhance the shared representation with useful information about superior global states. EA evolves the population using novel agent-level crossover and mutation operators, offering diverse experiences for MARL. Concurrently, MARL optimizes its policies and injects them into the population for evolution. The experiments on challenging continuous and discrete tasks demonstrate that RACE significantly improves the basic algorithms, consistently outperforming other algorithms. Our code is available at https://github.com/yeshenpy/RACE.

Recently, diffusion model shines as a promising backbone for the sequence modeling paradigm in offline reinforcement learning(RL). However, these works mostly lack the generalization ability across tasks with reward or dynamics change. To tackle this challenge, in this paper we propose a task-oriented conditioned diffusion planner for offline meta-RL(MetaDiffuser), which considers the generalization problem as conditional trajectory generation task with contextual representation. The key is to learn a context conditioned diffusion model which can generate task-oriented trajectories for planning across diverse tasks. To enhance the dynamics consistency of the generated trajectories while encouraging trajectories to achieve high returns, we further design a dual-guided module in the sampling process of the diffusion model. The proposed framework enjoys the robustness to the quality of collected warm-start data from the testing task and the flexibility to incorporate with different task representation method. The experiment results on MuJoCo benchmarks show that MetaDiffuser outperforms other strong offline meta-RL baselines, demonstrating the outstanding conditional generation ability of diffusion architecture.

Many score-based active learning methods have been successfully applied to graph-structured data, aiming to reduce the number of labels and achieve better performance of graph neural networks based on predefined score functions. However, these algorithms struggle to learn policy distributions that are proportional to rewards and have limited exploration capabilities. In this paper, we innovatively formulate the graph active learning problem as a generative process, named GFlowGNN, which generates various samples through sequential actions with probabilities precisely proportional to a predefined reward function. Furthermore, we propose the concept of flow nodes and flow features to efficiently model graphs as flows based on generative flow networks, where the policy network is trained with specially designed rewards. Extensive experiments on real datasets show that the proposed approach has good exploration capability and transferability, outperforming various state-of-the-art methods.

Multi-task reinforcement learning (MTRL) is a technique to train multiple tasks simultaneously, where previous works usually train a single model to solve different tasks by sharing parameters across various tasks. However, these methods are faced with inter-task interference since what parameters should be shared across tasks is not addressed, dramatically reducing learning efficiency. To solve these problems, we propose a novel MTRL framework called Task-Specific feature Selector and Scheduler (T3S), which consists of two components: a feature selector and a task scheduler. Specifically, the feature selectors employ hypernetworks to construct task-specific soft masks, which can be applied by globally shared representation to construct task-specific features. The task scheduler selects tasks for learning through two metrics, where the selection probability is inversely proportional to task progress (e.g., success rate) and task learning speed. Experimental results show that T3S consistently outperforms the state-of-the-art MTRL algorithms on various robotics manipulation tasks.

The automotive industry is shifting from hardware-centric to software-centric with the emergence of various intelligent features powered by software. This poses a new challenge for software testers to ensure software reliability by designing test plans that satisfy the test objectives while abiding by the constraints like scope, time, as well as various automotive safety standards. This paper proposed an automatic test plan generation framework built on the evolutionary algorithm. A novel encoding mechanism is proposed to represent the multi-dimensional test plan, while a belief model is proposed to reveal the underlying correlations between the relevant test attributes. Experiments conducted on an actual automotive software in production environment developed by our industry partner show that our method can achieve around 50% improvements in finding defects and covering high-priority test cases as compared to typical evolutionary algorithms while abiding by multiple constraints such as the total run time and custom objectives set by users.

Learning to collaborate is critical in Multi-Agent Reinforcement Learning (MARL). Previous works promote collaboration by maximizing the correlation of agents’ behaviors, which is typically characterized by Mutual Information (MI) in different forms. However, we reveal sub-optimal collaborative behaviors also emerge with strong correlations, and simply maximizing the MI can, surprisingly, hinder the learning towards better collaboration. To address this issue, we propose a novel MARL framework, called Progressive Mutual Information Collaboration (PMIC), for more effective MI-driven collaboration. PMIC uses a new collaboration criterion measured by the MI between global states and joint actions. Based on this criterion, the key idea of PMIC is maximizing the MI associated with superior collaborative behaviors and minimizing the MI associated with inferior ones. The two MI objectives play complementary roles by facilitating better collaborations while avoiding falling into sub-optimal ones. Experiments on a wide range of MARL benchmarks show the superior performance of PMIC compared with other algorithms.

Deep Reinforcement Learning (DRL) has been a promising solution to many complex decision-making problems. Nevertheless, the notorious weakness in generalization among environments prevent widespread application of DRL agents in real-world scenarios. Although advances have been made recently, most prior works assume sufficient online interaction on training environments, which can be costly in practical cases. To this end, we focus on an offline-training-online-adaptation setting, in which the agent first learns from offline experiences collected in environments with different dynamics and then performs online policy adaptation in environments with new dynamics. In this paper, we propose Policy Adaptation with Decoupled Representations (PAnDR) for fast policy adaptation. In offline training phase, the environment representation and policy representation are learned through contrastive learning and policy recovery, respectively. The representations are further refined by mutual information optimization to make them more decoupled and complete. With learned representations, a Policy-Dynamics Value Function (PDVF) network is trained to approximate the values for different combinations of policies and environments from offline experiences. In online adaptation phase, with the environment context inferred from few experiences collected in new environments, the policy is optimized by gradient ascent with respect to the PDVF. Our experiments show that PAnDR outperforms existing algorithms in several representative policy adaptation problems.

Despite achieving superior performance in human-level control problems, unlike humans, deep reinforcement learning (DRL) lacks high-order intelligence (e.g., logic deduction and reuse), thus it behaves ineffectively than humans regarding learning and generalization in complex problems. Previous works attempt to directly synthesize a white-box logic program as the DRL policy, manifesting logic-driven behaviors. However, most synthesis methods are built on imperative or declarative programming, and each has a distinct limitation, respectively. The former ignores the cause-effect logic during synthesis, resulting in low generalizability across tasks. The latter is strictly proof-based, thus failing to synthesize programs with complex hierarchical logic. In this paper, we combine the above two paradigms together and propose a novel Generalizable Logic Synthesis (GALOIS) framework to synthesize hierarchical and strict cause-effect logic programs. GALOIS leverages the program sketch and defines a new sketch-based hybrid program language for guiding the synthesis. Based on that, GALOIS proposes a sketch-based program synthesis method to automatically generate white-box programs with generalizable and interpretable cause-effect logic. Extensive evaluations on various decision-making tasks with complex logic demonstrate the superiority of GALOIS over mainstream baselines regarding the asymptotic performance, generalizability, and great knowledge reusability across different environments.

Exploring the underlying structure of a Human-Machine Interface (HMI) product effectively while adhering to the pre-defined test conditions and methodology is critical for validating the quality of the software. We propose an reinforcement-learning powered Automated Software Structure Exploration Framework for Testing (ASSET), which is capable of interacting with and analyzing the HMI software under testing (SUT). The main challenge is to incorporate the human instructions into the ASSET phase by using the visual feedback such as the downloaded image sequence from the HMI, which could be difficult to analyze. Our framework combines both computer vision and natural language processing techniques to understand the semantic meanings of the visual feedback. Building on the semantic understanding, we develop a rules-guided software exploration algorithm via reinforcement learning and deterministic finite automaton (DFA). We conducted experiments on HMI software in actual production phase and demonstrate that the exploration coverage and efficiency of our framework outperforms current start-of-art methods.

Despite the impressive success achieved in various domains, deep reinforcement learning (DRL) is still faced with the sample inefficiency problem. Transfer learning (TL), which leverages prior knowledge from different but related tasks to accelerate the target task learning, has emerged as a promising direction to improve RL efficiency. The majority of prior work considers TL across tasks with the same state-action spaces, while transferring across domains with different state-action spaces is relatively unexplored. Furthermore, such existing cross-domain transfer approaches only enable transfer from a single source policy, leaving open the important question of how to best transfer from multiple source policies. This paper proposes a novel framework called Cross-domain Adaptive Transfer (CAT) to accelerate DRL. CAT learns the state-action correspondence from each source task to the target task and adaptively transfers knowledge from multiple source task policies to the target policy. CAT can be easily combined with existing DRL algorithms and experimental results show that CAT significantly accelerates learning and outperforms other cross-domain transfer methods on multiple continuous action control tasks.

Web testing has long been recognized as a notoriously difficult task. Even nowadays, web testing still mainly relies on manual efforts in many cases while automated web testing is still far from achieving human-level performance. Key challenges include dynamic content update and deep bugs hiding under complicated user interactions and specific input values, which can only be triggered by certain action sequences in the huge space of all possible sequences. In this paper, we propose WebExplor, an automatic end-to-end web testing framework, to achieve an adaptive exploration of web applications. WebExplor adopts a curiosity-driven reinforcement learning to generate high-quality action sequences (test cases) with temporal logical relations. Besides, WebExplor incrementally builds an automaton during the online testing process, which acts as the high-level guidance to further improve the testing efficiency. We have conducted comprehensive evaluations on six real-world projects, a commercial SaaS web application, and performed an in-the-wild study of the top 50 web applications in the world. The results demonstrate that in most cases WebExplor can achieve significantly higher failure detection rate, code coverage and efficiency than existing state-of-the-art web testing techniques. WebExplor also detected 12 previously unknown failures in the commercial web application, which have been confirmed and fixed by the developers. Furthermore, our in-the-wild study further uncovered 3,466 exceptions and errors.

Agent-based negotiation aims at automating negotiation process on behalf of humans to save time and efforts. While successful, the research of automated negotiation focuses on communication merely through offer exchange (e.g., following alternating offer protocol). As many real-world settings involve linguistic channel to express intentions, ask questions, discuss plans and so on, the information bandwidth is therefore restricted and grounded in the action space of negotiation. To bridge the gap, this work proposes the MCAN (Multiple Channel Automated Negotiation) agent that is based on deep reinforcement learning (DRL). It is capable of negotiating using both offers and linguistic communication channels with autonomous agents or human players. Specifically, the agent leverages Parametrized Deep Q-Network (P-DQN) that provides solution for a hybrid discrete-continuous action space, thereby learning a comprehensive negotiation strategy which integrates linguistic communication skills and bidding strategies. The experimental results show that the MCAN agent can successfully complete negotiations with both proposal and linguistic communication, and it outperforms baseline agents in terms of averaged utility.

Discovering the underlying structure of HMI software efficiently and sufficiently for the purpose of testing without any prior knowledge on the software logic remains a difficult problem. The key challenge lies in the complexity of the HMI software and the high variance in the coverage of current methods. In this paper, we introduce the PathFinder, an effective and automatic HMI software exploration framework. PathFinder adopts a curiosity-based reinforcement learning framework to choose actions that lead to the discovery of more unknown states. Additionally, PathFinder progressively builds a navigation model during the exploration to further improve state coverage. We have conducted experiments on both simulations and real-world HMI software testing environment, which comprise a full tool chain of automobile dashboard instrument cluster. The exploration coverage outperforms manual and fuzzing methods which are the current industrial standards.

With the increasing popularity of electric vehicles, distributed energy generation and storage facilities in smart grid systems, an efficient Demand-Side Management (DSM) is urgent for energy savings and peak loads reduction. Traditional DSM works focusing on optimizing the energy activities for a single household can not scale up to large-scale home energy management problems. Multi-agent Deep Reinforcement Learning (MA-DRL) shows a potential way to solve the problem of scalability, where modern homes interact together to reduce energy consumers consumption while striking a balance between energy cost and peak loads reduction. However, it is difficult to solve such an environment with the non-stationarity, and existing MA-DRL approaches cannot effectively give incentives for expected group behavior. In this paper, we propose a collective MA-DRL algorithm with continuous action space to provide fine-grained control on a large scale microgrid. To mitigate the non-stationarity of the microgrid environment, a novel predictive model is proposed to measure the collective market behavior. Besides, a collective behavior entropy is introduced to reduce the high peak loads incurred by the collective behaviors of all householders in the smart grid. Empirical results show that our approach significantly outperforms the state-of-the-art methods regarding power cost reduction and daily peak loads optimization.

Adversarial attacks against conventional Deep Learning (DL) systems and algorithms have been widely studied, and various defenses were proposed. However, the possibility and feasibility of such attacks against Deep Reinforcement Learning (DRL) are less explored. As DRL has achieved great success in various complex tasks, designing effective adversarial attacks is an indispensable prerequisite towards building robust DRL algorithms. In this paper, we introduce two novel adversarial attack techniques to stealthily and efficiently attack the DRL agents. These two techniques enable an adversary to inject adversarial samples in a minimal set of critical moments while causing the most severe damage to the agent. The first technique is the critical point attack: the adversary builds a model to predict the future environmental states and agent's actions, assesses the damage of each possible attack strategy, and selects the optimal one. The second technique is the antagonist attack: the adversary automatically learns a domain-agnostic model to discover the critical moments of attacking the agent in an episode. Experimental results demonstrate the effectiveness of our techniques. Specifically, to successfully attack the DRL agent, our critical point technique only requires 1 (TORCS) or 2 (Atari Pong and Breakout) steps, and the antagonist technique needs fewer than 5 steps (4 Mujoco tasks), which are significant improvements over state-of-the-art methods.

Generating diverse behaviors for game artificial intelligence (Game AI) has been long recognized as a challenging task in the game industry. Designing a Game AI with a satisfying behavioral characteristic (style) heavily depends on the domain knowledge and is hard to achieve manually. Deep reinforcement learning sheds light on advancing the automatic Game AI design. However, most of them focus on creating a superhuman Game AI, ignoring the importance of behavioral diversity in games. To bridge the gap, we introduce a new framework, named EMOGI, which can automatically generate desirable styles with almost no domain knowledge. More importantly, EMOGI succeeds in creating a range of diverse styles, providing behavior-diverse Game AIs. Evaluations on the Atari and real commercial games indicate that, compared to existing algorithms, EMOGI performs better in generating diverse behaviors and significantly improves the efficiency of Game AI design.

Reinforcement learning agents usually learn from scratch, which requires a large number of interactions with the environment. This is quite different from the learning process of human. When faced with a new task, human naturally have the common sense and use the prior knowledge to derive an initial policy and guide the learning process afterwards. Although the prior knowledge may be not fully applicable to the new task, the learning process is significantly sped up since the initial policy ensures a quick-start of learning and intermediate guidance allows to avoid unnecessary exploration. Taking this inspiration, we propose knowledge guided policy network (KoGuN), a novel framework that combines human prior suboptimal knowledge with reinforcement learning. Our framework consists of a fuzzy rule controller to represent human knowledge and a refine module to finetune suboptimal prior knowledge. The proposed framework is end-to-end and can be combined with existing policy-based reinforcement learning algorithm. We conduct experiments on several control tasks. The empirical results show that our approach, which combines suboptimal human knowledge and RL, achieves significant improvement on learning efficiency of flat RL algorithms, even with very low-performance human prior knowledge.

Multiagent algorithms often aim to accurately predict the behaviors of other agents and find a best response accordingly. Previous works usually assume an opponent uses a stationary strategy or randomly switches among several stationary ones. However, an opponent may exhibit more sophisticated behaviors by adopting more advanced reasoning strategies, e.g., using a Bayesian reasoning strategy. This paper proposes a novel approach called Bayes-ToMoP which can efficiently detect the strategy of opponents using either stationary or higher-level reasoning strategies. Bayes-ToMoP also supports the detection of previously unseen policies and learning a best-response policy accordingly. We provide a theoretical guarantee of the optimality on detecting the opponent's strategies. We also propose a deep version of Bayes-ToMoP by extending Bayes-ToMoP with DRL techniques. Experimental results show both Bayes-ToMoP and deep Bayes-ToMoP outperform the state-of-the-art approaches when faced with different types of opponents in two-agent competitive games.

Smart grids are contributing to the demand-side management by integrating electronic equipment, distributed energy generation and storage and advanced meters and controllers. With the increasing adoption of electric vehicles and distributed energy generation and storage systems, residential energy management is drawing more and more attention, which is regarded as being critical to demand-supply balancing and peak load reduction. In this paper, we focus on a microgrid scenario in which modern homes interact together under a large-scale setting to better optimize their electricity cost. We first make households form a group with an economic stimulus. Then we formulate the energy expense optimization problem of the household community as a multi-agent coordination problem and present an Entropy-Based Collective Multiagent Deep Reinforcement Learning (EB-C-MADRL) framework to address it. Experiments with various real-world data demonstrate that EB-C-MADRL can reduce both the long-term group power consumption cost and daily peak demand effectively compared with existing approaches.

Game testing has been long recognized as a notoriously challenging task, which mainly relies on manual playing and scripting based testing in game industry. Even until recently, automated game testing still remains to be largely untouched niche. A key challenge is that game testing often requires to play the game as a sequential decision process. A bug may only be triggered until completing certain difficult intermediate tasks, which requires a certain level of intelligence. The recent success of deep reinforcement learning (DRL) sheds light on advancing automated game testing, without human competitive intelligent support. However, the existing DRLs mostly focus on winning the game rather than game testing. To bridge the gap, in this paper, we first perform an in-depth analysis of 1349 real bugs from four real-world commercial game products. Based on this, we propose four oracles to support automated game testing, and further propose Wuji, an on-the-fly game testing framework, which leverages evolutionary algorithms, DRL and multi-objective optimization to perform automatic game testing. Wuji balances between winning the game and exploring the space of the game. Winning the game allows the agent to make progress in the game, while space exploration increases the possibility of discovering bugs. We conduct a large-scale evaluation on a simple game and two popular commercial games. The results demonstrate the effectiveness of Wuji in exploring space and detecting bugs. Moreover, Wuji found 3 previously unknown bugs, which have been confirmed by the developers, in the commercial games.

In multiagent domains, coping with non-stationary agents that change behaviors from time to time is a challenging problem, where an agent is usually required to be able to quickly detect the other agent's policy during online interaction, and then adapt its own policy accordingly. This paper studies efficient policy detecting and reusing techniques when playing against non-stationary agents in Markov games. We propose a new deep BPR+ algorithm by extending the recent BPR+ algorithm with a neural network as the value-function approximator. To detect policy accurately, we propose the \textit{rectified belief model} taking advantage of the \textit{opponent model} to infer the other agent's policy from reward signals and its behaviors. Instead of directly storing individual policies as BPR+, we introduce \textit{distilled policy network} that serves as the policy library in BPR+, using policy distillation to achieve efficient online policy learning and reuse. Deep BPR+ inherits all the advantages of BPR+ and empirically shows better performance in terms of detection accuracy, cumulative rewards and speed of convergence compared to existing algorithms in complex Markov games with raw visual inputs.

A major challenge in document clustering is the extremely high dimensionality as well as the sparsity of the sample matrix. In this paper, we propose a new short-text oriented analysis approach to cluster short text automatically and extract the hot topics from each cluster. Different from the previous studies focused on long text, our analysis approach mainly focused on short-text cases. The approach consists of three stages: Firstly, generate feature vector for each sample so as to obtain the whole high-dimensional Vector Space Model; Secondly, use Singular Value Decomposition to achieve the dimensions reduction; Lastly, apply cosine similarity and k-means method to cluster samples on the low-dimensional matrix and extract the hot topics for each cluster. The experimental results show that our analysis approach can deal with the short-text samples and find out the hot topics efficiently and effectively.