Dynamic Risk-Aware MPPI for Mobile Robots in Crowds via Efficient Monte Carlo Approximations

Real-Robot Experiment with Pedestrians

A Clearpath Jackal using our approach navigates among pedestrians while minimizing its Collision Probability (CP) as well as mantaining the CP below a desired threshold. Pedestrians, tracked via motion capture, have their uncertain future motion predicted by a uni-modal Gaussian Distribution.

Abstract

Deploying mobile robots safely among humans requires the motion planner to account for the uncertainty in the other agents' predicted trajectories. This remains challenging in traditional approaches, especially with arbitrarily shaped predictions and real-time constraints. To address these challenges, we propose a Dynamic Risk-Aware Model Predictive Path Integral control (DRA-MPPI), a motion planner that incorporates uncertain future motions modelled with potentially non-Gaussian stochastic predictions. By leveraging MPPI’s gradient-free nature, we propose a method that efficiently approximates the joint Collision Probability (CP) among multiple dynamic obstacles for several hundred sampled trajectories in real-time via a Monte Carlo (MC) approach. This enables the rejection of samples exceeding a predefined CP threshold or the integration of CP as a weighted objective within the navigation cost function. Consequently, DRA-MPPI mitigates the freezing robot problem while enhancing safety. Real-world and simulated experiments with multiple dynamic obstacles demonstrate DRA-MPPI’s superior performance compared to state-of-the-art approaches, including Scenario-based Model Predictive Control (S-MPC), Frenét planner, and vanilla MPPI.

Comparisons with Non-Gaussian Predictions

Although not visualized in the video, all pedestrians are predicted to have a small chance of turning left any future time-step. This is modeled with a Gaussian Mixture Model. Below is a video comparison to a state-of-the-art Scenario-Based MPC [de Groot, 2023], more comparisons are provided in the tables and figures of the paper.

DRA-MPPI (Ours)
Scenario-Based MPC

Comparisons with Gaussian Predictions

Below are the video comparisons showing a Jackal robot navigating a corridor with an increasing number of pedestrians. Pedestrian motion is simulated with a social forces model, but are predicted by the robot by a Gaussian distribution centered on a constant velocity model. This mismatch is added to make the simulation more realistic, as the prediction will not perfectly match actual motion. Comparisons with several other approaches are presented in the tables and figures in our paper.

DRA-MPPI (Ours)

Four pedestrians

Scenario-Based MPC

Four pedestrians

Eight pedestrians

Eight pedestrians

Twelve pedestrians

Twelve pedestrians