Wayfarer discovers RL options from a single experience stream, SOTA on hard Atari games
MarlosCMachado · x · 2026-10-23
Researchers at UAlberta/Amii (Erik M. Lintunen, Marlos C. Machado) present Wayfarer, a general, domain-agnostic online deep RL agent that discovers options via Laplacian representation learning from high-dimensional observations, using only a single stream of experience.
- Existing option discovery methods are limited to simple domains, rely on handcrafted representations, or barely help; Wayfarer works online without them.
- The discovered options simultaneously improve exploration, accelerate credit assignment, and generalize to unseen settings.
- Wayfarer achieves state-of-the-art among single-stream agents on the hardest Atari 2600 games, with the largest gains on long-horizon exploration titles like Montezuma's Revenge and Private Eye.
More from Research
- From generating candidates to scaling discovery: the future of scientific ML — ctjlewis · 2026-10-23
- HF's DNA model evals: Carbon-Mid-300B nears Evo2 with just 160B training tokens — _lewtun · 2026-10-23
- Google Quantum AI and Kaggle launch verifiable quantum circuit compilation contest — MeganRisdal · 2026-10-23
- CMU and Microsoft Research convene 150+ leaders to define AI agent evaluation standards — steph_milani · 2026-10-23
- RFD4-Proteina's 'underrated' docs detail partial diffusion for protein redesign — AllThingsApx · 2026-10-23
- MPI-Sec PhD Thesis on Vulnerability Detection Benchmarks Includes Pre-LLM Classic "Top Score on the Wrong Exam" — mboehme_ · 2026-10-23