Research Analyzer
← Back ICRA 2026

Agile in the Face of Delay: Asynchronous End-To-End Learning for Real-World Aerial Navigation

Yude Li, Zhexuan Zhou, Huizhe Li, Youmin Gong, Jie Mei

PDF

AI summary

Key figure (auto-extracted from paper)
Decoupling perception and control with explicit delay encoding enables high-frequency, robust aerial navigation in cluttered real-world environments despite low-frequency sensor updates.
Asynchronous learning end-to-end navigation aerial robotics temporal encoding sim-to-real transfer reinforcement learning

Problem

Modern end-to-end navigation for aerial vehicles suffers from a temporal mismatch between high-frequency control loops and low-frequency perception streams, forcing synchronous models into low control rates and causing data staleness.

Approach

The authors propose an asynchronous reinforcement learning framework that decouples perception and control, using a Temporal Encoding Module to explicitly model data staleness and a two-stage curriculum learning strategy for stable training and zero-shot sim-to-real transfer.

Key results

  • Novel asynchronous end-to-end architecture with efficient LiDAR pseudo-image processing
  • Theoretically-grounded Temporal Encoding Module compensating for perception delay
  • Two-stage curriculum learning enabling stable training and zero-shot sim-to-real transfer
  • Sustained 100 Hz control rate with robust navigation in cluttered real-world environments

Why it matters

Enables computationally constrained aerial vehicles to achieve high-frequency, safe autonomous navigation in complex environments without relying on high-bandwidth sensors or heavy onboard processing.

Abstract

Robust autonomous navigation for Autonomous Aerial Vehicles (AAVs) in complex environments is a critical capability. However, modern end-to-end navigation faces a key challenge: the high-frequency control loop needed for agile flight conflicts with low-frequency perception streams, which are limited by sensor update rates and significant computational cost. This mismatch forces conventional synchronous models into undesirably low control rates. To resolve this, we propose an asynchronous reinforcement learning framework that decou- ples perception and control, enabling a high-frequency policy to act on the latest IMU state for immediate reactivity, while in- corporating perception features asynchronously. To manage the resulting data staleness, we introduce a theoretically-grounded Temporal Encoding Module (TEM) that explicitly conditions the policy on perception delays, a strategy complemented by a two-stage curriculum to ensure stable and efficient training. Validated in extensive simulations, our method was successfully deployed in zero-shot sim-to-real transfer on an onboard NUC, where it sustains a 100 Hz control rate and demonstrates robust, agile navigation in cluttered real-world environments. Our project details are available at https://hitsz-mas. github.io/Agile-Asynch-Nav/.

Index terms

Aerial Systems: Perception and Autonomy Reinforcement Learning Motion and Path Planning

Related papers