Wayfarer discovers options via Laplacian representations, hitting SOTA on hardest Atari games
MarlosCMachado · x · 2026-10-05
A new arXiv paper by Erik M. Lintunen and Marlos C. Machado introduces Wayfarer, a general, domain-agnostic online deep RL agent that discovers options via Laplacian representation learning from high-dimensional observations.
- Prior option discovery methods were limited to simple domains, relied on handcrafted representations, or offered little improvement over option-free learning.
- The discovered options simultaneously improve exploration, accelerate credit assignment, and generalize to unseen settings.
- Wayfarer achieves state-of-the-art performance among single-stream agents on the hardest Atari 2600 games, with the largest gains on long-horizon titles like Montezuma's Revenge and Private Eye.
- Notable result: in Freeway, the agent naturally learns to represent only what it controls (moving up) rather than the cars.
- Machado says it fulfills the promise of options first sketched in his 10-year-old paper "Learning Purposeful Behaviour in the Absence of Rewards."
Related event: DeepMind's Wayfarer Masters Hard Atari Games via Discovered Options(9 posts)→
More from Research
- Auditing multi-agent collusion talk slated for COLM 2025 workshop — nandofioretto · 2026-10-05
- Fine-tuned 350M model lifts PII removal from 89.7% to 99.7% — JosephJacks_ · 2026-10-05
- Wondersearch Claims It Beat Every Dense Embedding Model on SciFact — Researchers Skeptical — NirantK · 2026-10-05
- Mathematicians plus Meta's Muse Spark solve 6 open math problems, skeptics unmoved — YiMaTweets · 2026-10-05
- Comprehensive Triton GPU programming lecture: from H100 internals to FlashAttention — kalyan_kpl · 2026-10-05
- MICCAI 2026 Accepts 1,167 of 4,402 Papers; FAU Erlangen Lands 11 Plus Two Challenge Wins — maier_ak · 2026-10-05