Single-Instance Sampling for Computationally Efficient and Accurate Real-Time Task Space MPPI Control
Dongwhan Kim, Euncheol Im, Yujin Kim, Myo-Taeg Lim, Yisoo Lee
AI summary
Problem
High-frequency real-time model predictive control for robot manipulators is hindered by heavy computational burdens, as conventional sampling-based MPC methods require too many samples to maintain accuracy.
Approach
The authors introduce single-instance sampling and a dynamic time horizon within an MPPI framework, leveraging GPU parallelization to efficiently explore trajectory space without relying on predefined reference inputs.
Key results
- Achieves 0.298 ms average computation time per control cycle
- Supports a 2.55-second prediction horizon
- Demonstrates fastest computation among MPC-based task space controllers
- Validated on a 7-DoF robotic arm with high control accuracy
Why it matters
Enables precise, reactive, and constraint-aware manipulation for high-DoF robots in dynamic environments, advancing real-time robotic control.
Abstract
This study presents a model predictive path integral (MPPI) method capable of conducting high-frequency real-time model predictive control (MPC) for robot manipulators. Real- time MPC-based manipulation holds significant potential for con- trolling an end-effector precisely and reactively while satisfying various constraints in dynamic environments. However, the opti- mization under a complex robot model and various constraints imposes a heavy computational burden, hindering the realization of high-frequency updates. To address this challenge, we propose a single-instance sampling-based MPPI algorithm and dynamic time horizon to significantly reduce the computational burden while enhancing control performance. The performance and efficacy of the proposed method are verified through experiments conducted on a 7-degree-of-freedom robotic arm, along with comparative simulations and analysis.