Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Planning with Fast Collision Checks for High Speed Navigation

Craig Knuth, Cora Dimmig, Brian Bittner

TL;DR

This work addresses high-speed navigation in obstacle-rich environments by learning a multi-modal distribution of motion primitives with normalizing flows and accelerating collision checks via a mask-based prior rejection. The GenPlan framework samples diverse, expert-styled maneuvers in real time, retaining multiple homotopies while evaluating collisions only for top candidates. Empirical results show GenPlan matching MPPI in random clutter and substantially higher exit rates in cul-de-sacs, highlighting its advantage in multi-modal planning without reliance on seed trajectories. The authors also propose hardware experiments and statistical safety certification (EVT) to improve robustness and safety in practical deployments.

Abstract

Reasoning about large numbers of diverse plans to achieve high speed navigation in cluttered environments remains a challenge for robotic systems even in the case of perfect perceptual information. Often, this is tackled by methods that iteratively optimize around a prior seeded trajectory and consequently restrict to local optima. We present a novel planning method using normalizing flows (NFs) to encode expert-styled motion primitives. We also present an accelerated collision checking framework that enables rejecting samples from the prior distribution before running them through the NF model for rapid sampling of collision-free trajectories. The choice of an NF as the generator permits a flexible way to encode diverse multi-modal behavior distributions while maintaining a smooth relation to the input space which allows approximating collision checks on NF inputs rather than outputs. We show comparable performance to model predictive path integral control in random cluttered environments and improved exit rates in a cul-de-sac environment. We conclude by discussing our plans for future work to improve both safety and performance of our controller.

Generative Planning with Fast Collision Checks for High Speed Navigation

TL;DR

This work addresses high-speed navigation in obstacle-rich environments by learning a multi-modal distribution of motion primitives with normalizing flows and accelerating collision checks via a mask-based prior rejection. The GenPlan framework samples diverse, expert-styled maneuvers in real time, retaining multiple homotopies while evaluating collisions only for top candidates. Empirical results show GenPlan matching MPPI in random clutter and substantially higher exit rates in cul-de-sacs, highlighting its advantage in multi-modal planning without reliance on seed trajectories. The authors also propose hardware experiments and statistical safety certification (EVT) to improve robustness and safety in practical deployments.

Abstract

Reasoning about large numbers of diverse plans to achieve high speed navigation in cluttered environments remains a challenge for robotic systems even in the case of perfect perceptual information. Often, this is tackled by methods that iteratively optimize around a prior seeded trajectory and consequently restrict to local optima. We present a novel planning method using normalizing flows (NFs) to encode expert-styled motion primitives. We also present an accelerated collision checking framework that enables rejecting samples from the prior distribution before running them through the NF model for rapid sampling of collision-free trajectories. The choice of an NF as the generator permits a flexible way to encode diverse multi-modal behavior distributions while maintaining a smooth relation to the input space which allows approximating collision checks on NF inputs rather than outputs. We show comparable performance to model predictive path integral control in random cluttered environments and improved exit rates in a cul-de-sac environment. We conclude by discussing our plans for future work to improve both safety and performance of our controller.
Paper Structure (10 sections, 2 equations, 2 figures, 3 tables)

This paper contains 10 sections, 2 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Method
  5. Experimental Setup
  6. Results
  7. Conclusions and Future Work
  8. Bicycle Dynamics
  9. MPPI Implementation
  10. MPPI Performance on Cul-de-Sac

Figures (2)

  • Figure 1: Samples for a single planning iteration. The green square identifies the region of interest where masking is performed. Red trajectories are masked, blue are not. Notice that one blue primitive goes through the lower obstacle; this primitive would be checked for collision if it was low cost. Also notice the samples are diverse across the homotopies of the environment.
  • Figure 2: Example rollouts in cul-de-sac (left) and random (right) worlds. Solid black is the rollout by GenPlan with selected plans at each planning iteration shown in blue. Dashed black is the rollout by MPPI with selected plans at each planning iteration shown in green. GenPlan is capable of exiting the cul-de-sac environment at much higher rates than MPPI.