C-Free-Uniform: A Map-Conditioned Trajectory Sampler for Model Predictive Path Integral Control
Yukang Cao, Rahul Moorthy Mahesh, Oguzhan Goktug Poyrazoglu, Volkan Isler
AI summary
Problem
Standard sampling-based MPC methods rely on unimodal noise distributions that struggle with initialization and local minima, while existing uniform sampling strategies ignore environmental geometry, wasting samples on collisions in cluttered spaces.
Approach
The authors define C-Free-Uniformity as a geometric objective for uniformly sampling collision-free reachable states and train a U-Net-based neural network conditioned on local occupancy maps to output a control distribution that achieves it, integrating it into a scalable MPPI controller.
Key results
- Introduced C-Free-Uniformity and a supervised learning method for collision-free trajectory sampling
- Developed CFU-MPPI controller that scales fixed-velocity policies for variable-speed navigation
- Demonstrated robust generalization across varying velocities, time steps, and horizons
- Achieved 57.4% success rate in cluttered navigation tasks, outperforming standard MPPI (33.0%) with fewer samples
Why it matters
It enables more efficient and reliable autonomous navigation in complex environments by eliminating wasted collision samples and reducing dependency on trajectory initialization.
Abstract
Trajectory sampling is a key component of sampling-based control mechanisms. Trajectory samplers rely on control input samplers, which generate control inputs u from a distribution p(u|x) where x is the current state. We introduce the notion of Free Configuration Space Uniformity (C-Free- Uniform for short) which has two key features: (i) the gen- erated control input can be used to uniformly sample the free configuration space, and (ii) in contrast to previously introduced trajectory sampling mechanisms where the distribution p(u|x) is independent of the environment, C-Free-Uniform is explicitly conditioned on the current local map. Next, we integrate this sampler into a new Model Predictive Path Integral (MPPI) Controller, CFU-MPPI. Experiments show that CFU-MPPI out- performs existing methods in terms of success rate in challeng- ing navigation tasks in cluttered polygonal environments while requiring a much smaller sampling budget. Code: https: //github.com/ogpoyrazoglu/cuniform_sampling.