Selected work

Representative Papers

Green-VLA: Staged Vision-Language-Action Model for Generalist Robots

Jan 31, 2026

This work addresses the challenge of generalizing control and ensuring safe execution for general-purpose robots across heterogeneous morphologies—such as humanoids, mobile manipulators, and fixed-base arms—in real-world environments. The authors propose a staged vision-language-action framework that integrates a five-phase curriculum learning strategy, a unified embodied perception-action interface, pretraining with multimodal foundation models, and a safety-enhancement mechanism during inference. Innovatively combining reinforcement learning policy alignment, temporally aligned data processing, and out-of-distribution detection, the approach significantly improves task success rates, robustness, and long-horizon execution efficiency. Extensive experiments in both simulation and real-world robotic platforms demonstrate the method’s effectiveness and performance gains in cross-platform deployment.

0 citationsRead paper

Robust RL Control for Bipedal Locomotion with Closed Kinematic Chains

Jul 14, 2025

Conventional reinforcement learning (RL) control for bipedal robots with closed-chain kinematics often simplifies the structure to an open-chain model, leading to inaccurate modeling of joint coupling, friction dynamics, and motor-space characteristics—and consequently poor sim-to-real transfer performance. Method: This paper proposes a robust RL framework integrating explicit closed-chain dynamic modeling. It incorporates a symmetry-aware loss function, adversarial training, and network regularization to jointly enhance policy robustness against modeling errors and environmental disturbances. Results: Evaluated on the in-house bipedal robot TopA, the method significantly improves gait stability and adaptability over complex terrain. Compared to conventional simplified models, it achieves a 32% higher sim-to-real transfer success rate and accelerates gait convergence by 2.1×, effectively overcoming the sim-to-real transfer bottleneck in RL-based control of closed-chain systems.

0 citationsRead paper

Recent publications

Latest Papers

Green-VLA: Staged Vision-Language-Action Model for Generalist Robots

Jan 31, 2026

This work addresses the challenge of generalizing control and ensuring safe execution for general-purpose robots across heterogeneous morphologies—such as humanoids, mobile manipulators, and fixed-base arms—in real-world environments. The authors propose a staged vision-language-action framework that integrates a five-phase curriculum learning strategy, a unified embodied perception-action interface, pretraining with multimodal foundation models, and a safety-enhancement mechanism during inference. Innovatively combining reinforcement learning policy alignment, temporally aligned data processing, and out-of-distribution detection, the approach significantly improves task success rates, robustness, and long-horizon execution efficiency. Extensive experiments in both simulation and real-world robotic platforms demonstrate the method’s effectiveness and performance gains in cross-platform deployment.

0 citationsRead paper

Robust RL Control for Bipedal Locomotion with Closed Kinematic Chains

Jul 14, 2025

Conventional reinforcement learning (RL) control for bipedal robots with closed-chain kinematics often simplifies the structure to an open-chain model, leading to inaccurate modeling of joint coupling, friction dynamics, and motor-space characteristics—and consequently poor sim-to-real transfer performance. Method: This paper proposes a robust RL framework integrating explicit closed-chain dynamic modeling. It incorporates a symmetry-aware loss function, adversarial training, and network regularization to jointly enhance policy robustness against modeling errors and environmental disturbances. Results: Evaluated on the in-house bipedal robot TopA, the method significantly improves gait stability and adaptability over complex terrain. Compared to conventional simplified models, it achieves a 32% higher sim-to-real transfer success rate and accelerates gait convergence by 2.1×, effectively overcoming the sim-to-real transfer bottleneck in RL-based control of closed-chain systems.

0 citationsRead paper