A Reinforcement Learning Based Motion Planner for Quadrotor Autonomous Flight in Dense Environment

2024-08-01

arXiv.org

AI Summary

In dense, complex environments, conventional quadrotor trajectory optimization suffers from a decoupling between planning and control, underutilization of dynamic capabilities, sluggish response, and poor safety. Method: This paper proposes a hierarchical planning framework integrating visibility-graph-based path search with deep reinforcement learning (PPO). At the high level, a topology-feasible path is generated using visibility graphs and heuristic search; at the low level, PPO learns to generate high-dynamic, high-precision executable trajectories in real time, enabling tight coordination between path planning and motion control. Contribution/Results: Evaluated in indoor dense-scene simulations, the method significantly reduces flight time while improving path safety and dynamic adaptability over traditional static-time-allocation optimization approaches. It establishes a new paradigm for agile and robust autonomous flight.

Technology Category

• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Intelligent Robots: Motion and Path Planning

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Quadrotor motion planning is critical for autonomous flight in complex environments, such as rescue operations. Traditional methods often employ trajectory generation optimization and passive time allocation strategies, which can limit the exploitation of the quadrotor's dynamic capabilities and introduce delays and inaccuracies. To address these challenges, we propose a novel motion planning framework that integrates visibility path searching and reinforcement learning (RL) motion generation. Our method constructs collision-free paths using heuristic search and visibility graphs, which are then refined by an RL policy to generate low-level motion commands. We validate our approach in simulated indoor environments, demonstrating better performance than traditional methods in terms of time span.

Problem

• Quadcopter
• Path Planning
• Flight Efficiency

Innovation

• Path Planning
• Learning-based Control
• Efficient Flight Strategies