Research Analyzer
← Back ICRA 2026

Flow-Aided Flight through Dynamic Clutters from Point to Motion

Bowen Xu, Zexuan Yan, Minghao Lu, Xiyu Fan, Yi Luo, Youshen Lin, Zhiqiang Chen, Yeke Chen, Qiyuan Qiao, Peng Lu

PDF

AI summary

Key figure (auto-extracted from paper)
A flow-aided LiDAR representation enables reinforcement learning to train a quadrotor for safe, real-time navigation through dynamic clutter without explicit object detection.
Autonomous Flight Dynamic Obstacle Avoidance LiDAR Perception Reinforcement Learning Optical Flow End-to-End Navigation

Problem

Navigating dynamic cluttered environments is hindered by the computational burden, latency, and unreliability of explicit object detection, tracking, and prediction pipelines, which struggle with occlusions and compound errors.

Approach

Raw LiDAR point clouds are encoded into a fixed-shape distance map and fused with multi-frame optical flow features to create a change-aware representation, which trains an end-to-end reinforcement learning policy for direct acceleration control.

Key results

  • Fixed-shape, low-resolution distance map preserves obstacle details while reducing computation
  • Multi-frame point flow captures environmental dynamics without object tracking
  • Change-aware RL policy implicitly learns early avoidance maneuvers
  • High success rate and real-world quadrotor deployment verified in complex dynamic clutters

Why it matters

Provides a deployment-ready, object-free navigation framework that enhances real-time drone autonomy in unpredictable, cluttered environments.

Abstract

Challenges in traversing dynamic clutters lie mainly in the efficient perception of the environmental dynamics and the generation of evasive behaviors considering obstacle movement. Previous solutions have made progress in explicitly modeling the dynamic obstacle motion for avoidance, but this key de- pendency of decision-making is time-consuming and unreliable in highly dynamic scenarios with occlusions. On the contrary, without introducing object detection, tracking, and prediction, we empower the reinforcement learning (RL) with single LiDAR sensing to realize an autonomous flight system directly from point to motion. For exteroception, a depth sensing distance map achieving fixed-shape, low-resolution, and detail-safe is encoded from raw point clouds, and an environment change sensing point flow is adopted as motion features extracted from multi-frame observations. These two are integrated into a lightweight and easy-to-learn representation of complex dynamic environments. For action generation, the behavior of avoiding dynamic threats in advance is implicitly driven by the proposed change-aware sensing representation, where the policy optimization is indi- cated by the relative motion modulated distance field. With the deployment-friendly sensing simulation and dynamics model- free acceleration control, the proposed system shows a superior success rate and adaptability to alternatives, and the policy derived from the simulator can drive a real-world quadrotor with safe maneuvers.

Index terms

Aerial Systems: Perception and Autonomy Reinforcement Learning Autonomous Vehicle Navigation

Related papers