RoboCerebra: Benchmark for Long-Horizon Robotic Manipulation

RoboCerebra is a large-scale benchmark designed to evaluate robotic manipulation in long-horizon tasks. It shifts the focus of robotic imitation learning from fast, reactive System 1 behavior to slow, deliberate System 2 reasoning. The benchmark was authored by qiukingballball and last updated on Hugging Face in October 2025.

Use Cases

• Evaluating robotic manipulation performance based on long-horizon tasks.
• Training imitation learning models that incorporate System 2 reasoning.
• Benchmarking progress in robotic planning and decision-making.

Strengths

• Explicitly designed for evaluating long-horizon robotic tasks.
• Provides conversion scripts and evaluation code via a linked GitHub repository.
• Last updated on Hugging Face on 2025-10-19 08:51:02.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

qiukingballball on Hugging Face

Collection Method

Likely a curated collection of robotic manipulation demonstrations or simulations.

Freshness

Last updated 2025-10-19 08:51:02.