POGEMA: A Benchmark Platform for Cooperative Multi-Agent Navigation
Alexey Skrynnik, Anton Andreychuk, Anatolii Borzilov, Alexander Chernyavskiy, Konstantin Yakovlev, Aleksandr Panov
2024-07-23
Summary
This paper introduces POGEMA, a new platform designed to help researchers evaluate and compare different methods for multi-agent navigation tasks. It provides tools for learning, problem generation, and benchmarking to facilitate fair assessments of various approaches.
What's the problem?
In the field of multi-agent reinforcement learning (MARL), it has been difficult to compare traditional methods with newer learning-based approaches because there hasn't been a unified framework for evaluation. This makes it hard to determine which methods work best for tasks like robot navigation and obstacle avoidance, especially since many existing methods rely on classical techniques that are not easily comparable to learning-based ones.
What's the solution?
The authors developed POGEMA, which includes a fast learning environment, a generator for creating problem instances, a collection of predefined problems, a visualization toolkit, and a benchmarking tool for automated evaluations. They also established an evaluation protocol that defines key metrics, such as success rate and path length, allowing for fair comparisons between classical, learning-based, and hybrid methods. Their experiments demonstrate how POGEMA can effectively compare various state-of-the-art techniques in multi-agent navigation tasks.
Why it matters?
This research is important because it provides a structured way to evaluate different approaches to multi-agent navigation, which can lead to better solutions in robotics. By making it easier to compare methods, POGEMA can help advance the development of more efficient and effective systems for tasks like coordinating multiple robots in complex environments.
Abstract
Multi-agent reinforcement learning (MARL) has recently excelled in solving challenging cooperative and competitive multi-agent problems in various environments with, mostly, few agents and full observability. Moreover, a range of crucial robotics-related tasks, such as multi-robot navigation and obstacle avoidance, that have been conventionally approached with the classical non-learnable methods (e.g., heuristic search) is currently suggested to be solved by the learning-based or hybrid methods. Still, in this domain, it is hard, not to say impossible, to conduct a fair comparison between classical, learning-based, and hybrid approaches due to the lack of a unified framework that supports both learning and evaluation. To this end, we introduce POGEMA, a set of comprehensive tools that includes a fast environment for learning, a generator of problem instances, the collection of pre-defined ones, a visualization toolkit, and a benchmarking tool that allows automated evaluation. We introduce and specify an evaluation protocol defining a range of domain-related metrics computed on the basics of the primary evaluation indicators (such as success rate and path length), allowing a fair multi-fold comparison. The results of such a comparison, which involves a variety of state-of-the-art MARL, search-based, and hybrid methods, are presented.