Dynamically Feasible Trajectory Generation with Optimization-Embedded Networks for Autonomous Flight
Zhichao Han, Long Xu, Liuao Pei, Fei Gao
AI summary
Problem
Purely learning-based navigation lacks interpretability and strict dynamic constraint satisfaction, while traditional modular systems suffer from latency and poor sub-module integration.
Approach
A neural network extracts safe flight corridors from depth data, which are embedded as differentiable spatial constraints into a model-based optimization problem to compute optimal trajectories via backpropagation.
Key results
- Differentiable trajectory optimization layer enables end-to-end network training
- Regularized motion primitive library efficiently captures multimodal planning distributions
- Outperforms state-of-the-art methods in trajectory smoothness, success rate, and constraint satisfaction
- Validated via real-world high-speed autonomous flight in dense forest environments
Why it matters
Provides a scalable and interpretable bridge between perception and planning, advancing reliable high-speed autonomous navigation for UAVs.
Abstract
This paper aims to bridge perception and planning in navigation systems by learning optimal trajectories from depth information in an end-to-end fashion. However, using neural net- works as closed-box replacements for traditional modules risks scalability and adaptability. Moreover, such methods often fall short in sufficiently incorporating the robot’s dynamic constraints, resulting in trajectories that are either inadequately executable or unexpectedly aggressive, diverging from user expectations. In this paper, we fuse the benefits of conventional methods and neural networksbyintroducinganoptimization-embeddednetworkbased on a compact trajectory library. The network distills spatial con- straints, which are then applied to model-based spatial-temporal trajectory optimization problem, yielding feasible and optimal so- lutions. By making the optimization problem differentiable, our model seamlessly approximates the optimal trajectory. Addition- ally, the introduced regularized trajectory library permits efficient capture of the spatial distribution of optimal trajectories with minimal storage cost, safeguarding multimodal planning features. Benchmarking demonstrates the outstanding performance of our method in trajectory smoothness, success rate, and constraint satis- faction. Real-world flight experiments with an onboard computer showcase the autonomous quadrotor’s ability to navigate swiftly through dense forests.