Reducing the Communication of Distributed Model Predictive Control: Autoencoders and Formation Control

TL;DR

This work tackles the data-communication bottleneck in distributed model predictive control (DMPC) for formation control of nonholonomic robots by integrating an undercomplete autoencoder to compress each agent's candidate input trajectory before inter-agent transmission. The autoencoder-augmented DMPC preserves closed-loop performance while reducing per-message data, and it outperforms naïve horizon-shortening strategies. Training leverages a dataset of about 2000 two-robot scenarios, and the approach generalizes to larger teams and more challenging conditions, including plant-model mismatch. The method is validated through simulations and physically distributed hardware experiments, demonstrating real-time feasibility and robustness even when full communication cannot meet timing constraints.

Abstract

Communication remains a key factor limiting the applicability of distributed model predictive control (DMPC) in realistic settings, despite advances in wireless communication. DMPC schemes can require an overwhelming amount of information exchange between agents as the amount of data depends on the length of the predication horizon, for which some applications require a significant length to formally guarantee nominal asymptotic stability. This work aims to provide an approach to reduce the communication effort of DMPC by reducing the size of the communicated data between agents. Using an autoencoder, the communicated data is reduced by the encoder part of the autoencoder prior to communication and reconstructed by the decoder part upon reception within the distributed optimization algorithm that constitutes the DMPC scheme. The choice of a learning-based reduction method is motivated by structure inherent to the data, which results from the data's connection to solutions of optimal control problems. The approach is implemented and tested at the example of formation control of differential-drive robots, which is challenging for optimization-based control due to the robots' nonholonomic constraints, and which is interesting due to the practical importance of mobile robotics. The applicability of the proposed approach is presented first in form of a simulative analysis showing that the resulting control performance yields a satisfactory accuracy. In particular, the proposed approach outperforms the canonical naive way to reduce communication by reducing the length of the prediction horizon. Moreover, it is shown that numerical experiments conducted on embedded computation hardware, with real distributed computation and wireless communication, work well with the proposed way of reducing communication even in practical scenarios in which full communication fails.

Figures (15)

• Figure 1: Autoencoder as built into the distributed optimization algorithm.
• Figure 2: Exemplary differential-drive mobile robot
• Figure 3: Heatmap comparison of the distributions of $\bm u$ and $\bm\gamma$ for the same scenarios for the the first input values in the prediction horizon corresponding to the input of one node of the autoencoder. The darker the shade of a grid cell, the more samples it contains.
• Figure 4: Initial states of the training scenarios with the color representing the initial orientation in the interval $[\qty[parse-numbers=false]{-\pi}{\radian}, \qty[parse-numbers=false]{\pi}{\radian})$
• Figure 5: Heatmap of trainings data distribution over the horizon, where time steps in the prediction horizon increase from left to right and from top to bottom. Thus, the current input is depicted in the top left and the end of the prediction horizon in the bottom right corner. This corresponds to the input nodes of the autoencoder.