Model Predictive Path Integral Control using Covariance Variable Importance Sampling

About

In this paper we develop a Model Predictive Path Integral (MPPI) control algorithm based on a generalized importance sampling scheme and perform parallel optimization via sampling using a Graphics Processing Unit (GPU). The proposed generalized importance sampling scheme allows for changes in the drift and diffusion terms of stochastic diffusion processes and plays a significant role in the performance of the model predictive control algorithm. We compare the proposed algorithm in simulation with a model predictive control version of differential dynamic programming.

Grady Williams, Andrew Aldrich, Evangelos Theodorou · 2015

Related benchmarks

Task	Dataset	Result
Offline Goal-Conditioned Planning	D4RL Maze2D Single Goal v0	Average Score16.2	14
Offline Goal-Conditioned Planning	D4RL Maze2D Multi Goals v0	Score (umaze)41.2	7
Offline Planning	Maze2D U-Maze single-task D4RL	Normalized Avg Return33.2	6
Offline Planning	Maze2D Large single-task D4RL	Normalized Avg Return5.1	6
Multi-task Offline Planning	Multi2D U-Maze multi-task D4RL	Normalized Average Return41.2	5
Multi-task Offline Planning	Multi2D Medium multi-task D4RL	Normalized Average Return15.4	5
Multi-task Offline Planning	Multi2D Large multi-task D4RL	Normalized Avg Return8	5
Trajectory Optimization	synthetic box-avoidance scenario	Success Rate60	4
Multi-agent human-robot navigation	Crowd interaction trajectory data Multi-agent scenario real-world pedestrian (test)	Collision Rate34.4	4
Multi-arm cooperative formation control	MuJoCo Four-arm Simulation	Time per step (ms)259.2	4