Selected work

Representative Papers

Contact-Anchored Policies: Contact Conditioning Creates Strong Robot Utility Models

Feb 09, 2026

Language instructions are often too abstract to support robust robotic manipulation in complex physical interactions. To address this limitation, this work proposes replacing linguistic commands with spatial contact points as the conditioning signal for policy learning, constructing a modular utility model library, and integrating it with EgoGym—a lightweight simulation platform—to enable rapid real-to-sim closed-loop iteration. Using only 23 hours of demonstration data, the approach achieves out-of-the-box, zero-shot generalization across environments and robot embodiments on three fundamental manipulation tasks, outperforming state-of-the-art vision-language-action models by 56% in performance. The core innovation lies in the novel contact-point-conditioned policy architecture, which substantially enhances both generalization capability and deployment efficiency.

0 citationsRead paper

LLM-GROP: Visually grounded robot task and motion planning with large language models

Oct 01, 2025The international journal of robotics research

This work addresses ambiguous multi-object arrangement tasks in mobile manipulation (MoMa), such as “setting a dinner table,” where target configurations are underspecified. Method: We propose a Task and Motion Planning (TAMP) framework integrating Large Language Model (LLM)-driven commonsense reasoning with vision-guided base pose optimization. The LLM performs high-level task decomposition and generates semantically plausible object poses grounded in everyday knowledge; a vision module learns optimal robot base poses to ensure reachability and collision avoidance; task and motion planning execute interleaved to guarantee action feasibility. Contribution/Results: To our knowledge, this is the first TAMP approach to explicitly embed LLM-derived semantic commonsense into the planning pipeline and realize a closed-loop synergy among vision, language, and motion. Evaluated on long-horizon object rearrangement in both simulation and real-world settings, our method achieves an 84.4% success rate in physical experiments. User studies indicate performance comparable to human servers, significantly improving generalization to unspecified target configurations.

0 citationsRead paper

Cascaded Diffusion Models for Neural Motion Planning

May 21, 2025

To address the challenge of collision constraint violations in global motion planning for robots operating in complex, cluttered environments, this paper proposes a cascaded hierarchical diffusion model. The first stage generates a topologically feasible coarse global trajectory, while the second stage refines it locally using multi-scale features. An online collision detection and re-optimization mechanism is further integrated to strictly enforce global geometric constraints. This work represents the first application of diffusion-based policies to end-to-end collision-free global trajectory generation. Evaluated on navigation and dexterous manipulation tasks, the method achieves approximately 5% higher success rates compared to state-of-the-art baselines. It significantly improves trajectory safety, feasibility, and generalization across diverse environments and task configurations.

0 citationsRead paper

Exercise Specialists Evaluation of Robot-led Physical Therapy for People with Parkinsons Disease

Feb 07, 2025

This study addresses low adherence to home-based rehabilitation among Parkinson’s disease patients by evaluating the effectiveness and acceptability of a robot-led physical therapy system from the perspective of clinical exercise specialists (ESs). Using a mixed-methods approach—including the Technology Acceptance Model (TAM), NASA-TLX workload assessment, semi-structured interviews, and behavioral observation—the study systematically integrates ES input for the first time, yielding a human–robot collaborative rehabilitation design framework. Eleven ESs consistently reported that the system enhances adherence to home exercise, patient engagement, and training consistency. They further identified “naturalness of feedback” and “operational simplicity” as two critical dimensions for optimization. Innovatively, this work embeds expert clinical feedback early in the rehabilitation robot design loop. It thus establishes a reusable evaluation paradigm and actionable optimization pathway for embodied intelligent rehabilitation systems targeting neurodegenerative disorders.

0 citationsRead paper

Recent publications

Latest Papers

Contact-Anchored Policies: Contact Conditioning Creates Strong Robot Utility Models

Feb 09, 2026

Language instructions are often too abstract to support robust robotic manipulation in complex physical interactions. To address this limitation, this work proposes replacing linguistic commands with spatial contact points as the conditioning signal for policy learning, constructing a modular utility model library, and integrating it with EgoGym—a lightweight simulation platform—to enable rapid real-to-sim closed-loop iteration. Using only 23 hours of demonstration data, the approach achieves out-of-the-box, zero-shot generalization across environments and robot embodiments on three fundamental manipulation tasks, outperforming state-of-the-art vision-language-action models by 56% in performance. The core innovation lies in the novel contact-point-conditioned policy architecture, which substantially enhances both generalization capability and deployment efficiency.

0 citationsRead paper

LLM-GROP: Visually grounded robot task and motion planning with large language models

Oct 01, 2025The international journal of robotics research

This work addresses ambiguous multi-object arrangement tasks in mobile manipulation (MoMa), such as “setting a dinner table,” where target configurations are underspecified. Method: We propose a Task and Motion Planning (TAMP) framework integrating Large Language Model (LLM)-driven commonsense reasoning with vision-guided base pose optimization. The LLM performs high-level task decomposition and generates semantically plausible object poses grounded in everyday knowledge; a vision module learns optimal robot base poses to ensure reachability and collision avoidance; task and motion planning execute interleaved to guarantee action feasibility. Contribution/Results: To our knowledge, this is the first TAMP approach to explicitly embed LLM-derived semantic commonsense into the planning pipeline and realize a closed-loop synergy among vision, language, and motion. Evaluated on long-horizon object rearrangement in both simulation and real-world settings, our method achieves an 84.4% success rate in physical experiments. User studies indicate performance comparable to human servers, significantly improving generalization to unspecified target configurations.

0 citationsRead paper

Cascaded Diffusion Models for Neural Motion Planning

May 21, 2025

To address the challenge of collision constraint violations in global motion planning for robots operating in complex, cluttered environments, this paper proposes a cascaded hierarchical diffusion model. The first stage generates a topologically feasible coarse global trajectory, while the second stage refines it locally using multi-scale features. An online collision detection and re-optimization mechanism is further integrated to strictly enforce global geometric constraints. This work represents the first application of diffusion-based policies to end-to-end collision-free global trajectory generation. Evaluated on navigation and dexterous manipulation tasks, the method achieves approximately 5% higher success rates compared to state-of-the-art baselines. It significantly improves trajectory safety, feasibility, and generalization across diverse environments and task configurations.

0 citationsRead paper

Exercise Specialists Evaluation of Robot-led Physical Therapy for People with Parkinsons Disease

Feb 07, 2025

This study addresses low adherence to home-based rehabilitation among Parkinson’s disease patients by evaluating the effectiveness and acceptability of a robot-led physical therapy system from the perspective of clinical exercise specialists (ESs). Using a mixed-methods approach—including the Technology Acceptance Model (TAM), NASA-TLX workload assessment, semi-structured interviews, and behavioral observation—the study systematically integrates ES input for the first time, yielding a human–robot collaborative rehabilitation design framework. Eleven ESs consistently reported that the system enhances adherence to home exercise, patient engagement, and training consistency. They further identified “naturalness of feedback” and “operational simplicity” as two critical dimensions for optimization. Innovatively, this work embeds expert clinical feedback early in the rehabilitation robot design loop. It thus establishes a reusable evaluation paradigm and actionable optimization pathway for embodied intelligent rehabilitation systems targeting neurodegenerative disorders.

0 citationsRead paper