CoCo: Online Mixed-Integer Control via Supervised Learning

TL;DR

CoCo is a data-driven algorithm that uses neural networks to rapidly generate high-quality solutions for mixed-integer convex programs in robotics, enabling real-time control with significant speed improvements.

Contribution

The paper introduces CoCo, a two-stage neural network-based method that efficiently solves MICPs online by predicting logical strategies, improving speed and solution quality over existing heuristics.

Findings

CoCo achieves 10-100x faster solutions compared to traditional solvers.

It produces near-optimal solutions suitable for real-time robot control.

The approach generalizes well to various robotics planning problems.

Abstract

Many robotics problems, from robot motion planning to object manipulation, can be modeled as mixed-integer convex programs (MICPs). However, state-of-the-art algorithms are still unable to solve MICPs for control problems quickly enough for online use and existing heuristics can typically only find suboptimal solutions that might degrade robot performance. In this work, we turn to data-driven methods and present the Combinatorial Offline, Convex Online (CoCo) algorithm for quickly finding high quality solutions for MICPs. CoCo consists of a two-stage approach. In the offline phase, we train a neural network classifier that maps the problem parameters to a (logical strategy), which we define as the discrete arguments and relaxed big-M constraints associated with the optimal solution for that problem. Online, the classifier is applied to select a candidate logical strategy given new problem parameters; applying this logical strategy allows us to solve the original MICP as a convex optimization problem. We show through numerical experiments how CoCo finds near optimal solutions to MICPs arising in robot planning and control with 1 to 2 orders of magnitude solution speedup compared to other data-driven approaches and solvers.