BVER transfers the bidirectional expansion of RRT-Connect to curriculum generation for goal-conditioned RL.
Bidirectional start-goal curriculum accelerates sparse-reward reinforcement learning without demonstrations.
Expanding from both the goal and the initial state simultaneously, the method consistently outperforms single-sided curricula on navigation, locomotion, and manipulation tasks.
Big Tech
Juri Pfammatter · Kaixian Qu · Clemens Schwarke · Victor Klemm · Marco Hutter
ETH Zürich · NVIDIA
Research Digest··3 min read
The authors introduce BVER, a curriculum learning algorithm that grows intermediate start states from the goal and intermediate goals from the initial state distribution, biasing both toward unexplored regions.
Why this paper
From NVIDIA and ETH Zürich
In one line
Bidirectional Voronoi-biased exploration curricula, growing starts from goals and goals from starts, accelerate sparse-reward goal-conditioned reinforcement learning.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§