ManiSkill PushT: Robotic Manipulation Demonstrations

ManiSkill PushT Demonstrations is a collection of robotic manipulation trajectories. It contains approximately 1000 demonstrations generated via a PPO reinforcement learning policy. The dataset was authored by haosulab and last updated on August 19, 2024.

Use Cases

• Training imitation learning agents based on provided demonstration trajectories.
• Benchmarking reinforcement learning algorithms against a pre-collected demonstration dataset.
• Analyzing policy behavior and failure modes in a robotic pushing task.
• Developing behavior cloning models for robotic control.

Strengths

• Contains approximately 1000 demonstration trajectories, providing a substantial base for imitation learning.
• Demonstrations were generated via a trained PPO policy, suggesting a consistent and potentially high-performance source.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

haosulab

Collection Method

Generated via a PPO reinforcement learning policy.

Freshness

Last updated 2024-08-19 23:47:26; freshness should be verified.