# APACE: Agile and Perception-Aware Trajectory Generation for Quadrotor Flights

**Authors:** Xinyi Chen, Yichen Zhang, Boyu Zhou, Shaojie Shen  
**arXiv:** [2403.08365](https://arxiv.org/abs/2403.08365)  
**Date:** 2024-03-13  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2403.08365v1/

## Key points

- Focus on enhancing state estimation accuracy during maneuvers in perception-aware planning for quadrotor flights
- Importance of feature matchability among frames for estimation accuracy
- Development of APACE (Agile and Perception-Aware Trajectory Generation) framework by Xinyi Chen, Yichen Zhang, Boyu Zhou, and Shaojie Shen
- APACE considers feature matchability during trajectory planning to reduce visual-based estimator errors and meet constraints related to smoothness, safety, agility, and quadrotor dynamics
- Objective is to maximize the number of covisible features while ensuring small enough parallax angles
- Proposal of a differentiable and accurate visibility model for efficient optimization resolution in trajectory planning
- Validation experiments show significant improvement in state estimation accuracy with RMSE reduction by up to an order of magnitude compared to existing methods
- Plan to release source code for APACE to benefit the wider community interested in perception-aware planning for quadrotors

## Summary

In the field of perception-aware planning for quadrotor flights, there has been a focus on enhancing state estimation accuracy during maneuvers. However, one crucial factor that has often been overlooked is the feature matchability among frames, which significantly impacts estimation accuracy. To address this gap, a team of researchers including Xinyi Chen, Yichen Zhang, Boyu Zhou, and Shaojie Shen have developed APACE (Agile and Perception-Aware Trajectory Generation), a novel framework for aggressive quadrotor flight. APACE takes into consideration feature matchability during trajectory planning to generate a perception-aware trajectory that not only reduces the error of visual-based estimators but also meets constraints related to smoothness, safety, agility, and quadrotor dynamics. The key objective of the framework is to maximize the number of covisible features while ensuring small enough parallax angles. Additionally, the researchers propose a differentiable and accurate visibility model that allows for efficient optimization resolution by decomposing the trajectory planning problem. Validation experiments conducted in both a photorealistic simulator and real-world scenarios demonstrate that trajectories generated by APACE significantly improve state estimation accuracy. In fact, root mean square error (RMSE) was reduced by up to an order of magnitude compared to existing methods. The researchers plan to release the source code for APACE to benefit the wider community interested in perception-aware planning for quadrotors. This work was accepted by ICRA2024 and showcases how advanced trajectory generation techniques can enhance performance in agile flight scenarios while considering perception-related factors.