LLA-MPPI: Rapidly Adaptive Whole-body Control of Legged Robots with GPU-Accelerated Parallel Simulations

This research project investigates the implementation of Look-back and Look-ahead Adaptive Model Predictive Path Integral control (LLA-MPPI) for systems operating under highly dynamic and uncertain environments. Traditional model predictive control schemes often degrade when physical parameters deviate from nominal design conditions. To address this, the methodology leverages a digital twin framework consisting of a bank of GPU-accelerated parallel simulators, each representing distinct physical or structural hypotheses. By evaluating windowed prediction errors in real time, the control system dynamically selects the simulator that best matches the active physical dynamics. A whole-body MPPI planner then optimizes control trajectories based on the selected model. This approach eliminates the need for extensive offline training

and provides physically interpretable model selection. The project focuses on evaluating the control framework's performance across diverse transient tasks, analyzing its adaptability to sudden payload variations, structural degradation, and external disturbances. The research provides structuring and implementation guidance for deploying parallelized, multi-model predictive control strategies in complex, real-time digital twin applications.

Message: Hi FE, I need help with “LLA-MPPI: Rapidly Adaptive Whole-body Control of Legged Robots with GPU-Accelerated Parallel Simulations” in “Chemical Engineering”