Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SafeMove-RL: A Certifiable Reinforcement Learning Framework for Dynamic Motion Constraints in Trajectory Planning

Tengfei Liu, Haoyang Zhong, Jiazheng Hu, Tan Zhang

TL;DR

SafeMove-RL addresses local motion planning in densely dynamic and uncertain environments by coupling a model-constrained reinforcement learning policy with a dynamic safety margin framework. It uses generated local trajectories as the observation space and introduces dynamic gap analysis and an adaptive safety-assessment mechanism to enable reliable online replanning, supported by sequential experience replay. Empirical results in simulation and on a real robot show superior success rates, smoother trajectories, and improved computational efficiency compared with multiple baselines. The approach advances safe, real-time navigation in unstructured spaces with potential applications in service, security, and disaster-response robotics.

Abstract

This study presents a dynamic safety margin-based reinforcement learning framework for local motion planning in dynamic and uncertain environments. The proposed planner integrates real-time trajectory optimization with adaptive gap analysis, enabling effective feasibility assessment under partial observability constraints. To address safety-critical computations in unknown scenarios, an enhanced online learning mechanism is introduced, which dynamically corrects spatial trajectories by forming dynamic safety margins while maintaining control invariance. Extensive evaluations, including ablation studies and comparisons with state-of-the-art algorithms, demonstrate superior success rates and computational efficiency. The framework's effectiveness is further validated on both simulated and physical robotic platforms.

SafeMove-RL: A Certifiable Reinforcement Learning Framework for Dynamic Motion Constraints in Trajectory Planning

TL;DR

SafeMove-RL addresses local motion planning in densely dynamic and uncertain environments by coupling a model-constrained reinforcement learning policy with a dynamic safety margin framework. It uses generated local trajectories as the observation space and introduces dynamic gap analysis and an adaptive safety-assessment mechanism to enable reliable online replanning, supported by sequential experience replay. Empirical results in simulation and on a real robot show superior success rates, smoother trajectories, and improved computational efficiency compared with multiple baselines. The approach advances safe, real-time navigation in unstructured spaces with potential applications in service, security, and disaster-response robotics.

Abstract

This study presents a dynamic safety margin-based reinforcement learning framework for local motion planning in dynamic and uncertain environments. The proposed planner integrates real-time trajectory optimization with adaptive gap analysis, enabling effective feasibility assessment under partial observability constraints. To address safety-critical computations in unknown scenarios, an enhanced online learning mechanism is introduced, which dynamically corrects spatial trajectories by forming dynamic safety margins while maintaining control invariance. Extensive evaluations, including ablation studies and comparisons with state-of-the-art algorithms, demonstrate superior success rates and computational efficiency. The framework's effectiveness is further validated on both simulated and physical robotic platforms.
Paper Structure (20 sections, 14 equations, 7 figures, 2 tables)

This paper contains 20 sections, 14 equations, 7 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Non-Learning Method for Dynamic Obstacle Avoidance
  4. Learning Methods for Dynamic Obstacle Avoidance
  5. METHOD
  6. Motion Planning in Dynamic Environments
  7. Setting Up the Observation Space
  8. Reward Function Setting
  9. Policy Framework Settings
  10. EXPERIMENTS
  11. Training settings
  12. Training Environment
  13. Evaluation Metrics
  14. Ablation Baseline
  15. Ablation Experiment
  16. ...and 5 more sections

Figures (7)

  • Figure 1: Robots realize dynamic obstacle avoidance through learning of dynamic clearance and state parameters.
  • Figure 2: The overall framework of the proposed algorithm. The framework initially acquires environmental data, including point cloud occupancy information, target point direction, and corresponding velocity. This data is then used to derive DWA path information. The path information is processed to generate a dynamic boundary, which interacts with the DWA algorithm to accelerate path selection. The generated dynamic boundary and DWA path information are subsequently integrated into the deep reinforcement learning (DRL) framework to determine the angular and linear velocities required for the mobile robot.
  • Figure 3: illustrates the trajectory planning framework: the green line indicates the Dynamic Window Approach (DWA)-generated reference path, while the light-blue region defines the safety corridor boundaries. The red trajectory demonstrates the optimized path adhering to these constraints.
  • Figure 4: Policy Framework Settings.
  • Figure 5: Four simulation environments with color-coded components: yellow cylinders denote dynamic obstacles governed by ORCA algorithm, blue spheres represent predefined target points, and red rectangular areas indicate unified initial positions. Environmental complexity increases from (a) to (d) through incremental addition of dynamic obstacles (8, 12, 16, 20), while maintaining consistent geometric constraints for target/start configurations.
  • ...and 2 more figures