New MPC framework adapts to unpredictable system dynamics

Researchers have introduced T2S-MPC, a novel framework designed to enhance model predictive control (MPC) for systems with unpredictable time-varying dynamics. This approach adaptively learns a residual dynamics model online, integrating it with a nominal model for improved planning. By encoding temporal information and using a two-timescale update scheme, T2S-MPC can capture nonstationary dynamics while maintaining stable learning. Evaluations on a 2D quadrotor demonstrated superior control performance and robustness compared to existing methods. AI

IMPACT This new adaptive control framework could improve the performance and robustness of robotic systems operating in dynamic environments.

RANK_REASON This is a research paper describing a new control method. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• 2D quadrotor
• T2S-MPC

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →