Selected work

Representative Papers

High-Speed, All-Terrain Autonomy: Ensuring Safety at the Limits of Mobility

Mar 20, 2026

This work addresses the challenge of simultaneously achieving high-speed navigation and rollover safety in complex off-road environments, where existing autonomous driving systems often lack real-time, highly maneuverable trajectory planning capabilities. The authors propose a novel local trajectory planner based on model predictive control (MPC) that integrates an accurate vehicle dynamics model tailored for non-planar terrain with an energy-based constraint mechanism. This formulation effectively mitigates rollover risks induced by extreme events such as wheel lift-off. Real-time performance is ensured through GPGPU-accelerated parallel computation. Both theoretical analysis and experimental validation demonstrate that the proposed approach significantly reduces rollover incidents and enhances mission success rates in extreme off-road scenarios, outperforming state-of-the-art methods.

0 citationsRead paper

MA-VLCM: A Vision Language Critic Model for Value Estimation of Policies in Multi-Agent Team Settings

Mar 16, 2026

This work addresses the limitations of traditional centralized critics in multi-agent reinforcement learning, which suffer from low sample efficiency, poor generalization, and deployment challenges in resource-constrained heterogeneous robotic systems. The authors propose MA-VLCM, a novel framework that leverages a pre-trained vision-language model (VLM) as a training-free centralized critic to estimate state values by integrating natural language task descriptions, visual trajectories, and multi-agent states. This approach significantly improves sample efficiency and cross-environment generalization while enabling the generation of lightweight policies. Experimental results demonstrate that MA-VLCM achieves strong zero-shot return prediction performance in both in-distribution and out-of-distribution multi-agent scenarios and is compatible with various VLM backbones.

0 citationsRead paper

Multi-Agent Deep Reinforcement Learning Under Constrained Communications

Jan 22, 2026

This work addresses the scalability, robustness, and generalization limitations in multi-agent reinforcement learning that arise from reliance on global state information—particularly the fragility observed under dynamic team compositions or environmental changes. To overcome these challenges, the authors propose a fully decentralized coordination framework that eschews all privileged centralized information, relying instead solely on local observations and peer-to-peer multi-hop communication for collaborative decision-making. The key innovations include a Distributed Graph Attention Network (D-GAT) for implicit global state inference and a novel Distributed Graph Attention MAPPO (DG-MAPPO) algorithm based on local policies and value functions. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art CTDE approaches across multiple benchmarks—including StarCraftII, Google Research Football, and Multi-Agent MuJoCo—and is effective for both homogeneous and heterogeneous agent teams.

0 citationsRead paper

Implicit Dual-Control for Visibility-Aware Navigation in Unstructured Environments

Jul 06, 2025

To address occlusions and unobserved regions arising from limited field-of-view in unknown unstructured environments for autonomous ground vehicles, this paper proposes a perception–control coupled two-layer planning framework. The method dynamically models perceptual uncertainty and implicitly embeds it into the Variational Autoencoder–Model Predictive Path Integral (VA-MPPI) optimization, enabling joint optimization of safety and navigation performance without explicit exploration objectives. By integrating a visibility-aware perception model with stochastic sampling–based model predictive control (MPC), the framework achieves online trade-offs between exploration and exploitation. Experimental evaluation across diverse off-road scenarios demonstrates an 84% task success rate and zero collision rate, significantly outperforming conventional deterministic controllers. The key contribution lies in the implicit incorporation of perceptual uncertainty into MPPI-based trajectory optimization, thereby unifying perception-aware decision-making and control under uncertainty.

0 citationsRead paper

Recent publications

Latest Papers

High-Speed, All-Terrain Autonomy: Ensuring Safety at the Limits of Mobility

Mar 20, 2026

This work addresses the challenge of simultaneously achieving high-speed navigation and rollover safety in complex off-road environments, where existing autonomous driving systems often lack real-time, highly maneuverable trajectory planning capabilities. The authors propose a novel local trajectory planner based on model predictive control (MPC) that integrates an accurate vehicle dynamics model tailored for non-planar terrain with an energy-based constraint mechanism. This formulation effectively mitigates rollover risks induced by extreme events such as wheel lift-off. Real-time performance is ensured through GPGPU-accelerated parallel computation. Both theoretical analysis and experimental validation demonstrate that the proposed approach significantly reduces rollover incidents and enhances mission success rates in extreme off-road scenarios, outperforming state-of-the-art methods.

0 citationsRead paper

MA-VLCM: A Vision Language Critic Model for Value Estimation of Policies in Multi-Agent Team Settings

Mar 16, 2026

This work addresses the limitations of traditional centralized critics in multi-agent reinforcement learning, which suffer from low sample efficiency, poor generalization, and deployment challenges in resource-constrained heterogeneous robotic systems. The authors propose MA-VLCM, a novel framework that leverages a pre-trained vision-language model (VLM) as a training-free centralized critic to estimate state values by integrating natural language task descriptions, visual trajectories, and multi-agent states. This approach significantly improves sample efficiency and cross-environment generalization while enabling the generation of lightweight policies. Experimental results demonstrate that MA-VLCM achieves strong zero-shot return prediction performance in both in-distribution and out-of-distribution multi-agent scenarios and is compatible with various VLM backbones.

0 citationsRead paper

Multi-Agent Deep Reinforcement Learning Under Constrained Communications

Jan 22, 2026

This work addresses the scalability, robustness, and generalization limitations in multi-agent reinforcement learning that arise from reliance on global state information—particularly the fragility observed under dynamic team compositions or environmental changes. To overcome these challenges, the authors propose a fully decentralized coordination framework that eschews all privileged centralized information, relying instead solely on local observations and peer-to-peer multi-hop communication for collaborative decision-making. The key innovations include a Distributed Graph Attention Network (D-GAT) for implicit global state inference and a novel Distributed Graph Attention MAPPO (DG-MAPPO) algorithm based on local policies and value functions. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art CTDE approaches across multiple benchmarks—including StarCraftII, Google Research Football, and Multi-Agent MuJoCo—and is effective for both homogeneous and heterogeneous agent teams.

0 citationsRead paper

Implicit Dual-Control for Visibility-Aware Navigation in Unstructured Environments

Jul 06, 2025

To address occlusions and unobserved regions arising from limited field-of-view in unknown unstructured environments for autonomous ground vehicles, this paper proposes a perception–control coupled two-layer planning framework. The method dynamically models perceptual uncertainty and implicitly embeds it into the Variational Autoencoder–Model Predictive Path Integral (VA-MPPI) optimization, enabling joint optimization of safety and navigation performance without explicit exploration objectives. By integrating a visibility-aware perception model with stochastic sampling–based model predictive control (MPC), the framework achieves online trade-offs between exploration and exploitation. Experimental evaluation across diverse off-road scenarios demonstrates an 84% task success rate and zero collision rate, significantly outperforming conventional deterministic controllers. The key contribution lies in the implicit incorporation of perceptual uncertainty into MPPI-based trajectory optimization, thereby unifying perception-aware decision-making and control under uncertainty.

0 citationsRead paper