Large Neighborhood Search for Multi-Agent Task Assignment and Path Finding with Precedence Constraints

Researchers Viraj Parimi and Brian C. Williams have published a new paper tackling the complex problem of coordinating robot teams, known as Task Assignment and Path Finding with Precedence Constraints (TAPF-PC). This problem is critical for real-world applications like warehouse logistics and manufacturing, where robots must not only navigate without collisions but also decide which robot does which task and in what specific order. The traditional approach, MAPF-PC, assumes task assignments are fixed, which often leads to inefficient solutions. The new work introduces a more flexible, holistic optimization that considers assignment, routing, and task order simultaneously.

To solve this computationally intense problem, the team developed a 'large neighborhood search' algorithm. It starts with a feasible but potentially suboptimal plan (a seed solution) and then iteratively improves it. In each iteration, the algorithm selects a subset of tasks and agents, reassigns the tasks within this 'neighborhood,' and then repairs the paths to ensure no collisions and that all precedence rules are still met. This method allows the system to escape local optima and find significantly better overall plans. In extensive experiments across multiple benchmark scenarios, their best algorithm configuration improved solutions in 89.1% of test instances compared to the initial fixed-assignment seeds, proving the substantial gains from integrated optimization.