MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents
arXiv cs.AI · 10 Oct 2026, 02:37 AEST
MC-EvoBench is a new benchmark designed to evaluate self-improvement in Minecraft agents. It aims to distinguish between capability changes and carried-over resources through controlled testing.
Key facts
- A central challenge in embodied recursive self-improvement is enabling agents to convert interaction history into capabilities that persist beyond one world and transfer to later tasks.
- We construct MC-EvoBench as a controlled Minecraft benchmark with a fixed catalog of 19 registered evolution chains, including 12 primary evaluation chains and seven calibration or probe chains.
- The benchmark reveals a consistent direct-capability gap between controller and planner-backed variants, while related experience yields no stable transfer.
- Stage audits further show that early wood-collection and crafting bottlenecks often leave later capabilities unobserved.
Read aloud
MC-EvoBench is a new benchmark designed to evaluate self-improvement in Minecraft agents. It aims to distinguish between capability changes and carried-over resources through controlled testing. A central challenge in embodied recursive self-improvement is enabling agents to convert interaction history into capabilities that persist beyond one world and transfer to later tasks. We construct MC-EvoBench as a controlled Minecraft benchmark with a fixed catalog of nineteen registered evolution chains, including twelve primary evaluation chains and seven calibration or probe chains. The benchmark reveals a consistent direct-capability gap between controller and planner-backed variants, while related experience yields no stable transfer. Stage audits further show that early wood-collection and crafting bottlenecks often leave later capabilities unobserved.
Sources in this cluster
- MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents
arXiv cs.AI · 10 Oct 2026, 02:37 AEST