We introduce a novel framework with strong generalization capabilities for enabling humanoid robots to perform semantically grounded, context-aware, and physically realizable expressive behaviors, with the goal of enhancing end-user engagement in open-world human–robot interaction (HRI). To achieve this goal, we develop a new large language model–based human cognition module that interprets user dialogs to infer latent intent and generates high-level multi-modal behavior descriptions. These semantic representations are mapped to speech and motion outputs through a dual-stage embodied behavior generation pipeline. The pipeline consists of a shape adaptation module that maps human body motions into the robot’s kinematic space, followed by a motion retargeting module that generates executable joint trajectories under physical constraints. Additionally, the modular architecture enables seamless integration of state-of-the-art generative models and serves as a practical testbed for evaluating expressive behavior generation in real-world settings. We validate this system on a 58-DoF humanoid platform through both controlled video-based studies and live HRI experiments. The results show significant improvements over rule-based and handcrafted baselines in terms of expressiveness, behavioral appeal, and user engagement. This work helps bridging the gap between high-level language understanding and low-level robot control, thereby enabling scalable and human-aligned expressive behavior generation for embodied agents.

Enabling humanoid robots to perform long-horizon mobile manipulation planning in real-world environments based on embodied perception and comprehension abilities has been a longstanding challenge. With the recent rise of large language models (LLMs), there has been a notable increase in the development of LLM-based planners. These approaches either utilize human-provided textual representations of the real world or heavily depend on prompt engineering to extract such representations, lacking the capability to quantitatively understand the environment, such as determining the feasibility of manipulating objects. To address these limitations, we present the Instruction-Augmented Long-Horizon Planning (IALP) system, a novel framework that employs LLMs to generate feasible and optimal actions based on real-time sensor feedback, including grounded knowledge of the environment, in a closed-loop interaction. Distinct from prior works, our approach augments user instructions into PDDL problems by leveraging both the abstract reasoning capabilities of LLMs and grounding mechanisms. By conducting various real-world long-horizon tasks, each consisting of seven distinct manipulatory skills, our results demonstrate that the IALP system can efficiently solve these tasks with an average success rate exceeding 80%. Our proposed method can operate as a high-level planner, equipping robots with substantial autonomy in unstructured environments through the utilization of multi-modal sensor inputs.

The challenges inherent in long-horizon tasks in robotics persist due to the typical inefficient exploration and sparse rewards in traditional reinforcement learning approaches. To address these challenges, we have developed a novel algorithm, termed hlexplicit-implicit subgoal planning (EISP), designed to tackle long-horizon tasks through a divide-and-conquer approach. We utilize two primary criteria, feasibility and optimality, to ensure the quality of the generated subgoals. EISP consists of three components: a hybrid subgoal generator, a hindsight sampler, and a value selector. The hybrid subgoal generator uses an explicit model to infer subgoals and an implicit model to predict the final goal, inspired by way of human thinking that infers subgoals by using the current state and final goal as well as reason about the final goal conditioned on the current state and given subgoals. Additionally, the hindsight sampler selects valid subgoals from an offline dataset to enhance the feasibility of the generated subgoals. While the value selector utilizes the value function in reinforcement learning to filter the optimal subgoals from subgoal candidates. To validate our method, we conduct four long-horizon tasks in both simulation and the real world. The obtained quantitative and qualitative data indicate that our approach achieves promising performance compared to other baseline methods. These experimental results can be seen on the website https://sites.google.com/view/vaesi.

Understanding the environments through interactions has been one of the most important human intellectual activities in mastering unknown systems. Deep reinforcement learning (DRL) has already been known to achieve effective control through human-like exploration and exploitation in many applications. However, the opaque nature of deep neural network (DNN) often hides critical information about feature relevance to control, which is essential for understanding the target systems. In this article, a novel online feature selection framework, namely, the dual-world-based attentive feature selection (D-AFS), is first proposed to identify the contribution of the inputs over the whole control process. Rather than the one world used in most DRL, D-AFS has both the real world and its virtual peer with twisted features. The newly introduced attention-based evaluation (AR) module performs the dynamic mapping from the real world to the virtual world. The existing DRL algorithms, with slight modification, can learn in the dual world. By analyzing the DRL's response in the two worlds, D-AFS can quantitatively identify respective features' importance toward control. A set of experiments is performed on four classical control systems in OpenAI Gym. Results show that D-AFS can generate the same or even better feature combinations than the solutions provided by human experts and seven recent feature selection baselines. In all cases, the selected feature representations are closely correlated with the ones used by underlying system dynamic models.

Reducing sensor requirements while keeping optimal control performance is crucial to many industrial control applications to achieve robust, low-cost, and computation-efficient controllers. However, existing feature selection solutions for the typical machine learning domain can hardly be applied in the domain of control with changing dynamics. In this paper, a novel framework, namely the Dual-world embedded Attentive Feature Selection (D-AFS), can efficiently select the most relevant sensors for the system under dynamic control. Rather than the one world used in most Deep Reinforcement Learning (DRL) algorithms, D-AFS has both the real world and its virtual peer with twisted features. By analyzing the DRL's response in two worlds, D-AFS can quantitatively identify respective features' importance towards control. A well-known active flow control problem, cylinder drag reduction, is used for evaluation. Results show that D-AFS successfully finds an optimized five-probes layout with 18.7% drag reduction than the state-of-the-art solution with 151 probes and 49.2% reduction than five-probes layout by human experts. We also apply this solution to four OpenAI classical control cases. In all cases, D-AFS achieves the same or better sensor configurations than originally provided solutions. Results highlight, we argued, a new way to achieve efficient and optimal sensor designs for experimental or industrial systems. Our source codes are made publicly available at https://github.com/G-AILab/DAFSFluid.

Recent advances in Multi-agent Reinforcement Learning (MARL) have made it possible to implement various tasks in cooperative as well as competitive scenarios through trial and error, and deep neural networks. These successes motivate us to bring the mechanism of MARL into the Multi-agent Resilient Consensus (MARC) problem that studies the consensus problem in a network of agents with faulty ones. Relying on the natural characteristics of the system goal, the key component in MARL, reward function, can thus be directly constructed via the relative distance among agents. Firstly, we apply Deep Deterministic Policy Gradient (DDPG) on each single agent to train and learn adjacent weights of neighboring agents in a distributed manner, that we call Distributed-DDPG (D-DDPG), so as to minimize the weights from suspicious agents and eliminate the corresponding influences. Secondly, to get rid of neural networks and their time-consuming training process, a Q-learning based algorithm, called Q-consensus, is further presented by building a proper reward function and a credibility function for each pair of neighboring agents so that the adjacent weights can update in an adaptive way. The experimental results indicate that both algorithms perform well with appearance of constant and/or random faulty agents, yet the Q-consensus algorithm outperforms the faulty ones running D-DDPG. Compared to the traditional resilient consensus strategies, e.g., Weighted-Mean-Subsequence-Reduced (W-MSR) or trustworthiness analysis, the proposed Q-consensus algorithm has greatly relaxed the topology requirements, as well as reduced the storage and computation loads. Finally, a smart-car hardware platform consisting of six vehicles is used to verify the effectiveness of the Q-consensus algorithm by achieving resilient velocity synchronization.