The computational burden of model predictive control (MPC) limits its application on real-time systems, such as robots, and often requires the use of short prediction horizons. This not only affects the control performance, but also increases the difficulty of designing MPC cost functions that reflect the desired long-term objective. ZipMPC presents a novel method that imitates a long-horizon MPC behaviour by learning a compressed and context-dependent cost function for a short-horizon MPC. It improves performance over alternative methods, such as approximate explicit MPC and automatic cost parameter tuning in terms of optimizing the long-term objective; maintaining computational costs comparable to a short-horizon MPC; ensuring constraint satisfaction; and generalizing control behaviour to environments not observed during training.
ZipMPC imitates long-horizon MPC behaviour by learning a compressed and context-dependent cost for a short-horizon MPC. It thereby combines the faster inference times of explicit MPC with the generalization and constraint recognition capabilities of automatic cost parameter tuning.
Applying ZipMPC on hardware, it can be seen how it incorporates the additional contextual information to improve racing performance in comparison to a standard MPC controller of identical horizon length and closely imitates the performance of a long horizon MPC.
@article{zipmpc2025,
title={ZipMPC: Compressed Context-Dependent MPC Cost via Imitation Learning},
author={Anonymous},
journal={Under Review},
year={2025}
}