MAPF-FrozenLake Benchmark: Multi-Agent Pathfinding Tasks for RL

MAPF-FrozenLake Benchmark is an evaluation dataset for multi-agent reinforcement learning. It was created by LARK-Lab for a research paper on training environments designed by large language models. The dataset was last updated on June 18, 2026.

Use Cases

• Benchmarking multi-agent reinforcement learning algorithms based on the described coordination tasks.
• Evaluating agent performance under varying coordination requirements based on the wait-ratio threshold configs.
• Studying the scalability of pathfinding solutions based on the splits by agent count.

Strengths

• Provides three distinct configurations (benchmark_wr025, benchmark_wr050, benchmark_wr075) for controlled experimentation.
• Includes three splits per configuration based on agent count, allowing for scalability analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

LARK-Lab

Collection Method

Created as an evaluation benchmark for a specific research paper.

Freshness

Last updated 2026-06-18 15:02:31; freshness should be verified.