WZWei Zhu

Wei Zhu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Wei Zhu has a total of 23 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • EmoBipedNav: Emotion-Aware Social Navigation for Bipedal Robots With Deep Reinforcement Learning

    by Wei Zhu, Abirath Raju, Abdulaziz Shamsah, Anqi Wu, Seth Hutchinson, Ye Zhao

    This study presents an emotion-aware navigation framework—EmoBipedNav—using deep reinforcement learning (DRL) for bipedal robots walking in socially interactive environments. The inherent complex dynamics of bipedal robots challenge their safe maneuvering capabilities in dynamic environments. Furthermore, the intricacies of social interactions and cues such as emotions significantly compound the challenges of bipedal robot navigation. To address these coupled issues, we propose a two-stage pipeline that jointly considers the bipedal full-body dynamics and the complexities of socially aware navigation. More specifically, an emotion-integrated navigation policy is developed to balance safety, efficiency, and social courtesy by responding to human emotional cues. One key component of the policy is a novel representation of social environments using sequential LiDAR grid maps, from which we extract latent features, implicitly including collision regions, discomfort zones determined by emotions, social interactions, and the evolving dynamics of the navigation system including robot movements and pedestrian motions. To accounts for path tracking errors and locomotion constraints during social navigation, we present an end-to-end navigation system that incorporates full-order robot dynamics during training. Finally, extensive benchmarking and sim-to-real experiments demonstrate that our method outperforms both model-based planners and DRL baselines, and generalizes effectively to real-world environments.


  • Open Access

    Iterative Adversarial Learning With Chaser Agents for Time-Efficient Crowd-Aware Navigation

    by Tianjian Yuan, Wei Zhu, Mitsuhiro Hayashibe

    This letter addresses the challenge of safe and time-efficient crowd navigation for autonomous robots in dynamic environments. Existing methods struggle in scenarios with unpredictable or obstructive pedestrian behaviors. These limitations raise serious safety and efficiency concerns in real-world deployments. To improve navigation robustness and efficiency, we propose an adversarial deep reinforcement learning (DRL) framework that simulates competitive pedestrian behaviors through a chaser agent. In addition, we further introduce an ensemble agent that dynamically selects policies based on real-time observations, enhancing generalization across diverse scenarios. Extensive simulation results demonstrate enhanced navigation performance in terms of success rate and navigation efficiency when using our framework. Additionally, the ensemble agent further improves stability and overall generalizability across various environments. Real-world experiments validate the approach and demonstrate the potential for sim-to-real transfer.


  • Open Access

    A Systematic Review of Robot Topological Semantic Mapping and Localization for Spatial Understanding

    by Feroz Fernando, Wei Zhu, Atsushi Yamashita, Mitsuhiro Hayashibe

    World representation is a fundamental robotics problem. Topological semantic maps compress complex environmental data into lightweight graph structures enriched with semantic information, forming compact world models that support high-level reasoning and advanced autonomy. Despite their scalability advantages over metric-semantic maps, they remain comparatively understudied. This review provides a comprehensive Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-guided analysis of topological semantic maps, uniquely integrating their construction and localization aspects. The first part examines the map construction pipeline: semantic abstraction from sensor data, graph topology generation, map maintenance, and encoding. The second part covers localization strategies, including descriptor matching for place recognition, structural and semantic consistency checks for hypothesis validation, and probabilistic and geometric pose refinement. The paper concludes by discussing application advantages, persistent challenges, and future directions for spatial understanding AI.

  • Spatiotemporal Feature-Encoded Navigation for ASVs in Unpredictable Maritime Scenarios

    by Zhiting Yao, Xiyuan Chen, Mitsuhiro Hayashibe, Ninghui Xu, Tianjian Yuan, Wei Zhu

    The paper proposes a robust collision avoidance framework for USVs navigating high-risk maritime environments without prior assumptions about surrounding obstacles. Current research typically assumes that known static obstacles and dynamic obstacles move in simplified patterns, failing to account for real-world conditions where human-operated vessels exhibit unpredictable and high-dynamic behaviors. To address this, we develop an innovative spatiotemporal feature-encoded navigation policy built on the implicit quantile network. The policy's input is constructed using a sensor-based observation that integrates multiple sensors for environmental data collection, enabling a highly responsive and end-to-end navigation solution. Additionally, by employing rectangle convolutional kernels to extract spatial features and utilizing a continuous-time closed-form network to capture long-term motion dependencies, the proposed method effectively distinguishes between static and dynamic obstacles. The algorithm is compared with one reinforcement learning-based method and three classical methods. The results demonstrate its superior performance in navigating USVs through congested scenarios involving varying ocean flows, aggressive vessels, and static barriers.

  • Open Access

    Learning Perceptive Legged Robot Locomotion in the Real World: A Systematic Review

    by Irfan Tito Kurniawan, Wei Zhu, Dai Owaki, Mitsuhiro Hayashibe

    Perception is essential for legged locomotion as it enables robots to anticipate upcoming terrains and obstacles, facilitating adaptive traversal of challenging environments. Recent advancements in learning methodologies and legged locomotion have fostered research on the integration of perception into legged robot locomotion controllers, allowing them to operate effectively in a variety of real-world environments that are inaccessible to wheeled or tracked robots. This review aims to summarize recent developments in perceptive legged robot locomotion and identify potential future research directions. Specifically, we restrict our scope to studies employing learning-based techniques. We examine two primary topics: 1) the applications and capabilities afforded by perceptive locomotion and 2) the methodologies for integrating perception into locomotion systems. We survey the characteristics, progress, and challenges associated with each application and methodology as well as the decisions involved. Finally, we provide a comprehensive overview of the current challenges and potential future research perspectives in perceptive legged robot locomotion.


  • Learn to Teach: Sample-Efficient Privileged Learning for Humanoid Locomotion Over Real-World Uneven Terrain

    by Feiyang Wu, Xavier Nal, Jaehwi Jang, Wei Zhu, Zhaoyuan Gu, Anqi Wu, Ye Zhao

    Humanoid robots promise transformative capabilities for industrial and service applications. While recent advances in Reinforcement Learning (RL) yield impressive results in locomotion, manipulation, and navigation, the proposed methods typically require enormous simulation samples to account for real-world variability. This work proposes a novel one-stage training framework—Learn to Teach (L2T)—which unifies teacher and student policy learning. Our approach recycles simulator samples and synchronizes the learning trajectories through shared dynamics, significantly reducing sample complexities and training time while achieving state-of-the-art performance. Furthermore, we validate the RL variant (L2T-RL) through extensive simulations and hardware tests on the Digit robot, demonstrating zero-shot sim-to-real transfer and robust performance over 12+ diverse terrains without depth estimation modules.


  • Open Access

    Autonomous Navigation System in Pedestrian Scenarios Using a Dreamer-Based Motion Planner

    by Wei Zhu, Mitsuhiro Hayashibe

    Navigation among pedestrians is a crucial capability of service robots; however, it is a challenge to manage time-varying environments stably. Recent deep reinforcement learning (DRL)-based approaches to crowd navigation have yielded numerous promising applications. However, they rely heavily on initial imitation learning and colossal positive datasets. Moreover, the difficulties in accurately localizing robots, detecting and tracking humans, representing and generalizing reciprocal human relationships restrict their deployment in real-world problems. We propose a Dreamer-based motion planner for collision-free navigation in diverse pedestrian scenarios. Our RL framework can completely learn from zero experience via a model-based DRL. The robot and humans are first projected onto a map, which is subsequently decoded into low-dimensional latent state. A predictive dynamic model in the latent space is jointly created to efficiently optimize the navigation policy. Additionally, we leverage the techniques of system identification, domain randomization, clustering and LiDAR SLAM for practical deployment. Simulation ablations and real implementations demonstrate that our motion planner outperforms state-of-the-art methods, and that the navigation system can be physically implemented in the real world.


  • Open Access

    A Hierarchical Deep Reinforcement Learning Framework With High Efficiency and Generalization for Fast and Safe Navigation

    by Wei Zhu, Mitsuhiro Hayashibe

    We present a hierarchical deep reinforcement learning (DRL) framework with prominent sampling efficiency and sim-to-real transfer ability for fast and safe navigation: the low-level DRL policy enables the robot to move toward the target position and keep a safe distance to obstacles simultaneously; the high-level DRL policy is supplemented to further enhance the navigation safety. We select a waypoint located on the path from the robot to the ultimate goal as the subgoal to reduce the state space and avoid sparse reward. Moreover, the path is generated based on either a local or a global map, which can significantly improve the sampling efficiency, safety, and generalization ability of the proposed DRL framework. Additionally, a target-directed representation for the action space can be derived based on the subgoal to improve the motion efficiency and reduce the action space. In order to demonstrate the eminent sampling efficiency, motion performance, obstacle avoidance, and generalization ability of the proposed framework, we implement sufficient comparisons with the nonlearning navigation methods and DRL-based baselines, with videos, data, code, and other supplemental material shown on our website.


  • A Survey of Sim-to-Real Transfer Techniques Applied to Reinforcement Learning for Bioinspired Robots

    by Wei Zhu, Xian Guo, Dai Owaki, Kyo Kutsuzawa, Mitsuhiro Hayashibe

    The state-of-the-art reinforcement learning (RL) techniques have made innumerable advancements in robot control, especially in combination with deep neural networks (DNNs), known as deep reinforcement learning (DRL). In this article, instead of reviewing the theoretical studies on RL, which were almost fully completed several decades ago, we summarize some state-of-the-art techniques added to commonly used RL frameworks for robot control. We mainly review bioinspired robots (BIRs) because they can learn to locomote or produce natural behaviors similar to animals and humans. With the ultimate goal of practical applications in real world, we further narrow our review scope to techniques that could aid in sim-to-real transfer. We categorized these techniques into four groups: 1) use of accurate simulators; 2) use of kinematic and dynamic models; 3) use of hierarchical and distributed controllers; and 4) use of demonstrations. The purposes of these four groups of techniques are to supply general and accurate environments for RL training, improve sampling efficiency, divide and conquer complex motion tasks and redundant robot structures, and acquire natural skills. We found that, by synthetically using these techniques, it is possible to deploy RL on physical BIRs in actuality.