PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving
本文提出PlannerForge，一种LLM代理框架，用于解决自动驾驶系统场景测试流程中的模块化问题，通过统一框架覆盖从场景生成到系统评估的全过程。
0 citationsRead paper
Selected work
Representative Papers
Scene Graph-Driven Haptic Feedback for Safety Enhancement in Robotic Ophthalmic Surgery via Physically Simulated iOCT
本文通过构建基于场景图的触觉反馈系统，利用物理模拟iOCT解决机器人眼科手术中因缺乏触觉反馈导致的精度问题。
0 citationsRead paper
Control-Informed Constraint Adaptation in Minimum-Time Trajectory Planning for Autonomous Racing
This study addresses the performance limitations in autonomous racing trajectory planning caused by neglecting execution errors. We propose a control-aware online planning framework that integrates real-time tracking deviations into the planning layer. By dynamically adjusting spatial constraints and iteratively expanding the planning horizon, this approach enables adaptive constraint handling and compensates for cumulative errors, effectively overcoming the bottlenecks of traditional modular architectures. High-fidelity closed-loop simulations demonstrate that the proposed framework reduces lap time by 1.8 seconds while maintaining time optimality and safety guarantees. With a median computation time of only 25 ms, this method significantly enhances the vehicle's capability to operate at the limits of track performance.
0 citationsRead paper
From Continuous Design to Delay-Aware Discrete Synthesis: Guaranteed High-Bandwidth Joint Control for PMSM Drives
This study addresses the challenge of achieving high-bandwidth control across wide operating conditions in robotic PMSM drives compromised by communication and computational delays. We propose a task-aware discrete modeling framework integrated with direct PI controller synthesis to overcome the limitations of conventional continuous-time design. By explicitly incorporating delay dynamics, this approach analytically determines optimal sampling frequencies and controller gains, enabling direct discrete controller design with guaranteed theoretical performance. The proposed method significantly reduces both sampling frequency and DC-bus voltage requirements. Simulation results and experiments on custom-built joint actuators validate its real-time efficacy in embedded systems, establishing a novel paradigm for high-performance robotic joint actuation under latency constraints.
0 citationsRead paper
Recent publications
Latest Papers
PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving
本文提出PlannerForge，一种LLM代理框架，用于解决自动驾驶系统场景测试流程中的模块化问题，通过统一框架覆盖从场景生成到系统评估的全过程。
0 citationsRead paper
Scene Graph-Driven Haptic Feedback for Safety Enhancement in Robotic Ophthalmic Surgery via Physically Simulated iOCT
本文通过构建基于场景图的触觉反馈系统，利用物理模拟iOCT解决机器人眼科手术中因缺乏触觉反馈导致的精度问题。
0 citationsRead paper
Control-Informed Constraint Adaptation in Minimum-Time Trajectory Planning for Autonomous Racing
This study addresses the performance limitations in autonomous racing trajectory planning caused by neglecting execution errors. We propose a control-aware online planning framework that integrates real-time tracking deviations into the planning layer. By dynamically adjusting spatial constraints and iteratively expanding the planning horizon, this approach enables adaptive constraint handling and compensates for cumulative errors, effectively overcoming the bottlenecks of traditional modular architectures. High-fidelity closed-loop simulations demonstrate that the proposed framework reduces lap time by 1.8 seconds while maintaining time optimality and safety guarantees. With a median computation time of only 25 ms, this method significantly enhances the vehicle's capability to operate at the limits of track performance.
0 citationsRead paper
From Continuous Design to Delay-Aware Discrete Synthesis: Guaranteed High-Bandwidth Joint Control for PMSM Drives
This study addresses the challenge of achieving high-bandwidth control across wide operating conditions in robotic PMSM drives compromised by communication and computational delays. We propose a task-aware discrete modeling framework integrated with direct PI controller synthesis to overcome the limitations of conventional continuous-time design. By explicitly incorporating delay dynamics, this approach analytically determines optimal sampling frequencies and controller gains, enabling direct discrete controller design with guaranteed theoretical performance. The proposed method significantly reduces both sampling frequency and DC-bus voltage requirements. Simulation results and experiments on custom-built joint actuators validate its real-time efficacy in embedded systems, establishing a novel paradigm for high-performance robotic joint actuation under latency constraints.
0 citationsRead paper