Jianguo Zhang, Thai Quoc Hoang, Ming Zhu, Zuxin Liu, Shiyu Wang, Tulika Manoj Awalgaonkar, Akshara Prabhakar, Haolin Chen, Weiran Yao, Zhiwei Liu, Juntao Tan, Juan Carlos Niebles, Shelby Heinecke, Huan Wang, Silvio Savarese, Caiming Xiong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Large language model (LLM) agents have shown great potential in solving real-world software engineering (SWE) problems. The most advanced open-source SWE agent can resolve over 27% of real GitHub issues in SWE-Bench Lite. However, these sophisticated agent frameworks exhibit varying strengths, excelling in certain tasks while underperforming in others. To fully harness the diversity of these agents, we propose DEI (Diversity Empowered Intelligence), a framework that leverages their unique expertise. DEI functions as a meta-module atop existing SWE agent frameworks, managing agent collectives for enhanced problem-solving. Experimental results show that a DEI-guided committee of agents is able to surpass the best individual agent's performance by a large margin. For instance, a group of open-source SWE agents, with a maximum individual resolve rate of 27.3% on SWE-Bench Lite, can achieve a 34.3% resolve rate with DEI, making a 25% improvement and beating most closed-source solutions. Our best-performing group excels with a 55% resolve rate, securing the highest ranking on SWE-Bench Lite. Our findings contribute to the growing body of research on collaborative AI systems and their potential to solve complex software engineering challenges.

Jianguo Zhang, Tian Lan, Ming Zhu, Zuxin Liu, Thai Quoc Hoang, Shirley Kokane, Weiran Yao, Juntao Tan, Akshara Prabhakar, Haolin Chen, Zhiwei Liu, Yihao Feng, Tulika Manoj Awalgaonkar, Rithesh R N, Zeyuan Chen, Ran Xu, Juan Carlos Niebles, Shelby Heinecke, Huan Wang, Silvio Savarese, Caiming Xiong. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Reinforcement learning (RL) has emerged as a powerful post-training technique to incentivize the reasoning ability of large language models (LLMs). However, LLMs can respond very inconsistently to RL finetuning: some show substantial performance gains, while others plateau or even degrade. To understand this divergence, we analyze the per-step influence of the RL objective and identify two key conditions for effective post-training: (1) RL-informative rollout accuracy, and (2) strong data co-influence, which quantifies how much the training data affects performance on other samples. Guided by these insights, we propose behavior injection, a task-agnostic data augmentation scheme applied prior to RL. Behavior injection enriches the supervised finetuning (SFT) data by seeding exploratory and exploitative behaviors, effectively making the model more RL-ready. We evaluate our method across two reasoning benchmarks with multiple base models. The results demonstrate that our theoretically motivated augmentation can significantly increase the performance gain from RL over the pre-RL model.

Training effective AI agents for multi-turn interactions requires high-quality data that captures realistic human-agent dynamics, yet such data is scarce and expensive to collect manually. We introduce APIGen-MT, a two-phase framework that generates verifiable and diverse multi-turn agent data. In the first phase, our agentic pipeline produces detailed task blueprints with ground-truth actions, leveraging a committee of LLM reviewers and iterative feedback loops. These blueprints are then transformed into complete interaction trajectories through simulated human-agent interplay. We train a family of models---the xLAM-2-fc-r series with sizes ranging from 1B to 70B parameters. Our models outperform frontier models such as GPT-4o and Claude 3.5 on $\tau$-bench and BFCL benchmarks, with the smaller models surpassing their larger counterparts, particularly in multi-turn settings, while maintaining superior consistency across multiple trials. Comprehensive experiments demonstrate that our verified blueprint-to-details approach yields high-quality training data, enabling the development of more reliable, efficient, and capable agents. We open-source both the synthetic data collected and the trained xLAM-2-fc-r models to advance research in AI agents.Dataset: https://huggingface.co/datasets/Salesforce/APIGen-MT-5k & Models: https://huggingface.co/collections/Salesforce/xlam-2-67ef5be12949d8dcdae354c4

Deep learning-based visual perception models lack robustness when faced with camera motion perturbations in practice. The current certification process for assessing robustness is costly and time-consuming due to the extensive number of image projections required for Monte Carlo sampling in the 3D camera motion space. To address these challenges, we present a novel, efficient, and practical framework for certifying the robustness of 3D-2D projective transformations against camera motion perturbations. Our approach leverages a smoothing distribution over the 2D-pixel space instead of in the 3D physical space, eliminating the need for costly camera motion sampling and significantly enhancing the efficiency of robustness certifications. With the pixel-wise smoothed classifier, we are able to fully upper bound the projection errors using a technique of uniform partitioning in camera motion space. Additionally, we extend our certification framework to a more general scenario where only a single-frame point cloud is required in the projection oracle. Through extensive experimentation, we validate the trade-off between effectiveness and efficiency enabled by our proposed method. Remarkably, our approach achieves approximately 80% certified accuracy while utilizing only 30% of the projected image frames.

Offline reinforcement learning (RL) offers a promising direction for learning policies from pre-collected datasets without requiring further interactions with the environment. However, existing methods struggle to handle out-of-distribution (OOD) extrapolation errors, especially in sparse reward or scarce data settings. In this paper, we propose a novel training algorithm called Conservative Density Estimation (CDE), which addresses this challenge by explicitly imposing constraints on the state-action occupancy stationary distribution. CDE overcomes the limitations of existing approaches, such as the stationary distribution correction method, by addressing the support mismatch issue in marginal importance sampling. Our method achieves state-of-the-art performance on the D4RL benchmark. Notably, CDE consistently outperforms baselines in challenging tasks with sparse rewards or insufficient data, demonstrating the advantages of our approach in addressing the extrapolation error problem in offline RL.

The full potential of large pretrained models remains largely untapped in control domains like robotics. This is mainly because of the scarcity of data and the computational challenges associated with training or fine-tuning these large models for such applications. Prior work mainly emphasizes either effective \emph{pretraining} of large models for decision-making or single-task adaptation. But real-world problems will require data-efficient, \emph{continual adaptation} for new control tasks. Recognizing these constraints, we introduce TAIL (Task-specific Adapters for Imitation Learning), a framework for efficient adaptation to new control tasks. Inspired by recent advancements in parameter-efficient fine-tuning in language domains, we explore efficient fine-tuning techniques---e.g., Bottleneck Adapters, P-Tuning, and Low-Rank Adaptation (LoRA)---in TAIL to adapt large pretrained models for new tasks with limited demonstration data. Our extensive experiments comparing prevalent parameter-efficient fine-tuning techniques and adaptation baselines suggest that TAIL with LoRA can achieve the best post-adaptation performance with only 1\% of the trainable parameters of full fine-tuning, while avoiding catastrophic forgetting and preserving adaptation plasticity in continual learning settings.

In the field of safe reinforcement learning (RL), finding a balance between satisfying safety constraints and optimizing reward performance presents a significant challenge. A key obstacle in this endeavor is the estimation of safety constraints, which is typically more difficult than estimating a reward metric due to the sparse nature of the constraint signals. To address this issue, we introduce a novel framework named Feasibility Consistent Safe Reinforcement Learning (FCSRL). This framework combines representation learning with feasibility-oriented objectives to identify and extract safety-related information from the raw state for safe RL. Leveraging self-supervised learning techniques and a more learnable safety metric, our approach enhances the policy learning and constraint estimation. Empirical evaluations across a range of vector-state and image-based tasks demonstrate that our method is capable of learning a better safety-aware embedding and achieving superior performance than previous representation learning baselines.

Cameras and LiDARs are both important sensors for autonomous driving, playing critical roles in 3D object detection. Camera-LiDAR Fusion has been a prevalent solution for robust and accurate driving perception. In contrast to the vast majority of existing arts that focus on how to improve the performance of 3D target detection through cross-modal schemes, deep learning algorithms, and training tricks, we devote attention to the impact of sensor configurations on the performance of learning-based methods. To achieve this, we propose a unified information-theoretic surrogate metric for camera and LiDAR evaluation based on the proposed sensor perception model. We also design an accelerated high-quality framework for data acquisition, model training, and performance evaluation that functions with the CARLA simulator. To show the correlation between detection performance and our surrogate metrics, We conduct experiments using several camera-LiDAR placements and parameters inspired by selfdriving companies and research institutions. Extensive experimental results of representative algorithms on nuScenes dataset validate the effectiveness of our surrogate metric, demonstrating that sensor configurations significantly impact point-cloudimage fusion based detection models, which contribute up to 30% discrepancy in terms of the average precision.

Safe Reinforcement Learning (RL) plays an important role in applying RL algorithms to safety-critical real-world applications, addressing the trade-off between maximizing rewards and adhering to safety constraints. This work introduces a novel approach that combines RL with trajectory optimization to manage this trade-off effectively. Our approach embeds safety constraints within the action space of a modified Markov Decision Process (MDP). The RL agent produces a sequence of actions that are transformed into safe trajectories by a trajectory optimizer, thereby effectively ensuring safety and increasing training stability. This novel approach excels in its performance on challenging Safety Gym tasks, achieving significantly higher rewards and near-zero safety violations during inference. The method’s real-world applicability is demonstrated through a safe and effective deployment in a real robot task of box-pushing around obstacles. Further insights are available from the videos and appendix on our website: https://sites.google.com/view/safemdp.

The advancement of function-calling agent models requires diverse, reliable, and high-quality datasets. This paper presents APIGen, an automated data generation pipeline designed to synthesize high-quality datasets for function-calling applications. We leverage APIGen and collect 3,673 executable APIs across 21 different categories to generate diverse function-calling datasets in a scalable and structured manner. Each data in our dataset is verified through three hierarchical stages: format checking, actual function executions, and semantic verification, improving its reliability and correctness. We demonstrate that models trained with our curated datasets, even with only 7B parameters, can achieve state-of-the-art performance on the Berkeley Function-Calling Benchmark, outperforming multiple GPT-4 models. Moreover, our 1B model achieves exceptional performance, surpassing GPT-3.5-Turbo and Claude-3 Haiku. We release a dataset containing 60,000 high-quality entries, aiming to advance the field of function-calling agent domains. The dataset and models are available on the project homepage \url{https://apigen-pipeline.github.io/}.

Previous work demonstrates that the optimal safe reinforcement learning policy in a noise-free environment is vulnerable and could be unsafe under observational attacks. While adversarial training effectively improves robustness and safety, collecting samples by attacking the behavior agent online could be expensive or prohibitively dangerous in many applications. We propose the robuSt vAriational ofF-policy lEaRning (SAFER) approach, which only requires benign training data without attacking the agent. SAFER obtains an optimal non-parametric variational policy distribution via convex optimization and then uses it to improve the parameterized policy robustly via supervised learning. The two-stage policy optimization facilitates robust training, and extensive experiments on multiple robot platforms show the efficiency of SAFER in learning a robust and safe policy: achieving the same reward with much fewer constraint violations during training than on-policy baselines.

Safe reinforcement learning (RL) trains a constraint satisfaction policy by interacting with the environment. We aim to tackle a more challenging problem: learning a safe policy from an offline dataset. We study the offline safe RL problem from a novel multi-objective optimization perspective and propose the $\epsilon$-reducible concept to characterize problem difficulties. The inherent trade-offs between safety and task performance inspire us to propose the constrained decision transformer (CDT) approach, which can dynamically adjust the trade-offs during deployment. Extensive experiments show the advantages of the proposed method in learning an adaptive, safe, robust, and high-reward policy. CDT outperforms its variants and strong offline safe RL baselines by a large margin with the same hyperparameters across all tasks, while keeping the zero-shot adaptation capability to different constraint thresholds, making our approach more suitable for real-world RL under constraints.

Different environments pose a great challenge to the outdoor robust visual perception for long-term autonomous driving, and the generalization of learning-based algorithms on different environments is still an open problem. Although monocular depth prediction has been well studied recently, few works focus on the robustness of learning-based depth prediction across different environments, e.g. changing illumination and seasons, owing to the lack of such a multi-environment real-world dataset and benchmark. To this end, the cross-season monocular depth prediction dataset and benchmark, SeasonDepth, is introduced to benchmark the depth estimation performance under different environments. We investigate several state-of-the-art representative open-source supervised and self-supervised depth prediction methods using newly-formulated metrics. Through extensive experimental evaluation on the proposed dataset and cross-dataset evaluation with current autonomous driving datasets, the performance and robustness against the influence of multiple environments are analyzed qualitatively and quantitatively. We show that long-term monocular depth prediction is still challenging and believe our work can boost further research on the long-term robustness and generalization for outdoor visual perception. The dataset is available on https://seasondepth.github.io.

Regardless of the particular task we want to perform in an environment, there are often shared safety constraints we want our agents to respect. For example, regardless of whether it is making a sandwich or clearing the table, a kitchen robot should not break a plate. Manually specifying such a constraint can be both time-consuming and error-prone. We show how to learn constraints from expert demonstrations of safe task completion by extending inverse reinforcement learning (IRL) techniques to the space of constraints. Intuitively, we learn constraints that forbid highly rewarding behavior that the expert could have taken but chose not to. Unfortunately, the constraint learning problem is rather ill-posed and typically leads to overly conservative constraints that forbid all behavior that the expert did not take. We counter this by leveraging diverse demonstrations that naturally occur in multi-task setting to learn a tighter set of constraints. We validate our method with simulation experiments on high-dimensional continuous control tasks.

Safe reinforcement learning (RL) focuses on training reward-maximizing agents subject to pre-defined safety constraints. Yet, learning versatile safe policies that can adapt to varying safety constraint requirements during deployment without retraining remains a largely unexplored and challenging area. In this work, we formulate the versatile safe RL problem and consider two primary requirements: training efficiency and zero-shot adaptation capability. To address them, we introduce the Conditioned Constrained Policy Optimization (CCPO) framework, consisting of two key modules: (1) Versatile Value Estimation (VVE) for approximating value functions under unseen threshold conditions, and (2) Conditioned Variational Inference (CVI) for encoding arbitrary constraint thresholds during policy optimization. Our extensive experiments demonstrate that CCPO outperforms the baselines in terms of safety and task performance while preserving zero-shot adaptation capabilities to different constraint thresholds data-efficiently. This makes our approach suitable for real-world dynamic applications.

The past few years have witnessed an increasing interest in improving the perception performance of LiDARs on au-tonomous vehicles. While most of the existing works focus on developing new deep learning algorithms or model ar-chitectures, we study the problem from the physical design perspective, i.e., how different placements of multiple Li-DARs influence the learning-based perception. To this end, we introduce an easy-to-compute information-theoretic sur-rogate metric to quantitatively and fast evaluate LiDAR placement for 3D detection of different types of objects. We also present a new data collection, detection model training and evaluation framework in the realistic CARLA simula-tor to evaluate disparate multi-LiDAR configurations. Using several prevalent placements inspired by the designs of self-driving companies, we show the correlation between our surrogate metric and object detection performance of different representative algorithms on KITTI through exten-sive experiments, validating the effectiveness of our LiDAR placement evaluation approach. Our results show that sen-sor placement is non-negligible in 3D point cloud-based ob-ject detection, which will contribute to 5% ~ 10% performance discrepancy in terms of average precision in chal-lenging 3D object detection settings. We believe that this is one of the first studies to quantitatively investigate the influence of LiDAR placement on perception performance.

Safe reinforcement learning (RL) aims to learn policies that satisfy certain constraints before deploying them to safety-critical applications. Previous primal-dual style approaches suffer from instability issues and lack optimality guarantees. This paper overcomes the issues from the perspective of probabilistic inference. We introduce a novel Expectation-Maximization approach to naturally incorporate constraints during the policy learning: 1) a provable optimal non-parametric variational distribution could be computed in closed form after a convex optimization (E-step); 2) the policy parameter is improved within the trust region based on the optimal variational distribution (M-step). The proposed algorithm decomposes the safe RL problem into a convex optimization phase and a supervised learning phase, which yields a more stable training performance. A wide range of experiments on continuous robotic tasks shows that the proposed method achieves significantly better constraint satisfaction performance and better sample efficiency than baselines. The code is available at https://github.com/liuzuxin/cvpo-safe-rl.

As shown by recent studies, machine intelligence-enabled systems are vulnerable to test cases resulting from either adversarial manipulation or natural distribution shifts. This has raised great concerns about deploying machine learning algorithms for real-world applications, especially in safety-critical domains such as autonomous driving (AD). On the other hand, traditional AD testing on naturalistic scenarios requires hundreds of millions of driving miles due to the high dimensionality and rareness of the safety-critical scenarios in the real world. As a result, several approaches for autonomous driving evaluation have been explored, which are usually, however, based on different simulation platforms, types of safety-critical scenarios, scenario generation algorithms, and driving route variations. Thus, despite a large amount of effort in autonomous driving testing, it is still challenging to compare and understand the effectiveness and efficiency of different testing scenario generation algorithms and testing mechanisms under similar conditions. In this paper, we aim to provide the first unified platform SafeBench to integrate different types of safety-critical testing scenarios, scenario generation algorithms, and other variations such as driving routes and environments. In particular, we consider 8 safety-critical testing scenarios following National Highway Traffic Safety Administration (NHTSA) and develop 4 scenario generation algorithms considering 10 variations for each scenario. Meanwhile, we implement 4 deep reinforcement learning-based AD algorithms with 4 types of input (e.g., bird’s-eye view, camera) to perform fair comparisons on SafeBench. We find our generated testing scenarios are indeed more challenging and observe the trade-off between the performance of AD agents under benign and safety-critical testing scenarios. We believe our unified platform SafeBench for large-scale and effective autonomous driving testing will motivate the development of new testing scenario generation and safe AD algorithms. SafeBench is available at https://safebench.github.io.

Safety is a critical concern when deploying reinforcement learning agents for realistic tasks. Recently, safe reinforcement learning algorithms have been developed to optimize the agent’s performance while avoiding violations of safety constraints. However, few studies have addressed the nonstationary disturbances in the environments, which may cause catastrophic outcomes. In this paper, we propose the context-aware safe reinforcement learning (CASRL) method, a metal-earning framework to realize safe adaptation in non-stationary environments. We use a probabilistic latent variable model to achieve fast inference of the posterior environment transition distribution given the context data. Safety constraints are then evaluated with uncertainty-aware trajectory sampling. Prior safety constraints are formulated with domain knowledge to improve safety during exploration. The algorithm is evaluated in realistic safety-critical environments with non-stationary disturbances. Results show that the proposed algorithm significantly outperforms existing baselines in terms of safety and robustness.

Multi-agent navigation in dynamic environments is of great industrial value when deploying a large scale fleet of robot to real-world applications. This paper proposes a decentralized partially observable multi-agent path planning with evolutionary reinforcement learning (MAPPER) method to learn an effective local planning policy in mixed dynamic environments. Reinforcement learning-based methods usually suffer performance degradation on long-horizon tasks with goal-conditioned sparse rewards, so we decompose the long-range navigation task into many easier sub-tasks under the guidance of a global planner, which increases agents' performance in large environments. Moreover, most existing multi-agent planning approaches assume either perfect information of the surrounding environment or homogeneity of nearby dynamic agents, which may not hold in practice. Our approach models dynamic obstacles' behavior with an image-based representation and trains a policy in mixed dynamic environments without homogeneity assumption. To ensure multi-agent training stability and performance, we propose an evolutionary training approach that can be easily scaled to large and complex environments. Experiments show that MAPPER is able to achieve higher success rates and more stable performance when exposed to a large number of non-cooperative dynamic obstacles compared with traditional reaction-based planner LRA* and the state-of-the-art learning-based method.

Continuously learning to solve unseen tasks with limited experience has been extensively pursued in meta-learning and continual learning, but with restricted assumptions such as accessible task distributions, independently and identically distributed tasks, and clear task delineations. However, real-world physical tasks frequently violate these assumptions, resulting in performance degradation. This paper proposes a continual online model-based reinforcement learning approach that does not require pre-training to solve task-agnostic problems with unknown task boundaries. We maintain a mixture of experts to handle nonstationarity, and represent each different type of dynamics with a Gaussian Process to efficiently leverage collected data and expressively model uncertainty. We propose a transition prior to account for the temporal dependencies in streaming data and update the mixture online via sequential variational inference. Our approach reliably handles the task distribution shift by generating new models for never-before-seen dynamics and reusing old models for previously seen dynamics. In experiments, our approach outperforms alternative methods in non-stationary tasks, including classic control with changing dynamics and decision making in different driving scenarios.

Autonomous vehicle manufacturers recognize that LiDAR provides accurate 3D views and precise distance measures under highly uncertain driving conditions. Its practical implementation, however, remains costly. This paper investigates the optimal LiDAR configuration problem to achieve utility maximization. We use the perception area and non-detectable subspace to construct the design procedure as solving a min-max optimization problem and propose a bio-inspired measure - volume to surface area ratio (VSR) - as an easy-to-evaluate cost function representing the notion of the size of the non-detectable subspaces of a given configuration. We then adopt a cuboid-based approach to show that the proposed VSR-based measure is a well-suited proxy for object detection rate. It is found that the Artificial Bee Colony evolutionary algorithm yields a tractable cost function computation. Our experiments highlight the effectiveness of our proposed VSR measure in terms of cost-effectiveness configuration as well as providing insightful analyses that can improve the design of AV systems.

Simultaneous Localization and Mapping (SLAM) is considered to be a fundamental capability for intelligent mobile robots. Over the past decades, many impressed SLAM systems have been developed and achieved good performance under certain circumstances. However, some problems are still not well solved, for example, how to tackle the moving objects in the dynamic environments, how to make the robots truly understand the surroundings and accomplish advanced tasks. In this paper, a robust semantic visual SLAM towards dynamic environments named DS-SLAM is proposed. Five threads run in parallel in DS-SLAM: tracking, semantic segmentation, local mapping, loop closing and dense semantic map creation. DS-SLAM combines semantic segmentation network with moving consistency check method to reduce the impact of dynamic objects, and thus the localization accuracy is highly improved in dynamic environments. Meanwhile, a dense semantic octo-tree map is produced, which could be employed for high-level tasks. We conduct experiments both on TUM RGB-D dataset and in real-world environment. The results demonstrate the absolute trajectory accuracy in DS-SLAM can be improved one order of magnitude compared with ORB-SLAM2. It is one of the state-of-the-art SLAM systems in high-dynamic environments.