PHYRE: A New Benchmark for Physical Reasoning
This provides a new benchmark for the AI/ML community to develop sample-efficient agents for physical reasoning, though it is incremental as it builds on existing benchmark concepts.
The authors tackled the problem of physical reasoning in intelligent agents by developing the PHYRE benchmark, a set of 2D classical mechanics puzzles, and found that modern learning algorithms fall short in solving them efficiently.
Understanding and reasoning about physics is an important ability of intelligent agents. We develop the PHYRE benchmark for physical reasoning that contains a set of simple classical mechanics puzzles in a 2D physical environment. The benchmark is designed to encourage the development of learning algorithms that are sample-efficient and generalize well across puzzles. We test several modern learning algorithms on PHYRE and find that these algorithms fall short in solving the puzzles efficiently. We expect that PHYRE will encourage the development of novel sample-efficient agents that learn efficient but useful models of physics. For code and to play PHYRE for yourself, please visit https://player.phyre.ai.