Event
HKU and Tsinghua Propose Offline Value Learning Framework VAST
1 reports · 1 sources
A research team from the University of Hong Kong and Tsinghua University proposed an offline value learning framework, VAST (Horizon Adaptive Offline Policy Learning via Value Stitching), which jointly selects future subgoals and time spans, stitching local returns with subsequent values, enabling offline reinforcement learning to adaptively choose the time scale for value updates. The research team evaluated the framework on OGBench, covering 10 domains and 50 state-input-based tasks. The current experimental results mainly come from state-input-based offline benchmarks; whether it can be extended to visual inputs, real robots, and more open environments remains to be verified.
Report timeline
- HKU and Tsinghua propose VAST: enabling offline reinforcement learning to adaptively select value update timescales
The University of Hong Kong and Tsinghua University propose VAST (Horizon Adaptive Offline Policy Learning via Value Stitching), an offline value learning framework that adaptively selects update horizons by jointly choosing future subgoals and time spans to stitch local returns with subsequent values.
机器之心 · 微信公众号 ·