openai/Video-Pre-Training
Video-Pre-Training
Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos
Builder
OpenAI
openai • ai-lab
Stars
1,674
Using upstream star count
Forks
169
Using upstream fork count
Open Issues
0
Activity Score
0/100
0 commits in 30d
Created
Jun 22, 2022
Project creation date
README Summary
Video PreTraining (VPT) is a semi-supervised imitation learning method that learns to act in complex environments by watching unlabeled online videos. The approach uses a large dataset of Minecraft gameplay videos to train agents that can perform sophisticated tasks without explicit demonstrations. VPT demonstrates that internet-scale video data can be leveraged to learn behavioral priors for reinforcement learning agents.
AI Dev Skills
Unmapped
Tags
Taxonomy
Deployment Context
Industries
Modalities
Skill Areas
Recent Activity
Updated 7 months ago
7 Days
0
30 Days
0
90 Days
0
Quality
research- Quality
- high
- Maturity
- research
Categories
PM Skills
Languages
Timeline
- Project created
- Jun 22, 2022
- Forked
- Mar 14, 2026
- Your last push
- 7 months ago
- Upstream last push
- 7 months ago
- Tracked since
- Sep 3, 2025
Similar Repos
pgvector cosine similarity · $0
Loading…