Wayfarer auto-discovers options to crack hard Atari games, beating DreamerV3 and Rainbow
MarlosCMachado · x · 2026-10-05
DeepMind researcher Marlos Machado shares a new RL method, Wayfarer, presented in Mastering Atari 2600 Games with Discovered Options.
- A general-purpose option-discovery method that works tabula rasa on high-dimensional observations from a single stream of experience, with no domain knowledge
- Learns an agent-centric Laplacian representation whose intrinsic reward shapes options, which in turn shape experience in a virtuous cycle
- Simultaneously improves exploration, credit assignment, and transfer, outperforming DreamerV3, Rainbow, and IQN
- Targets historically hard Atari games where difficult exploration and credit assignment slowed progress
Related event: DeepMind's Wayfarer Masters Hard Atari Games via Discovered Options(9 posts)→
More from Research
- Next step for FlowBank: let the workflow bank evolve through deployment — furongh · 2026-10-05
- FlowBank: NeurIPS paper reuses complementary agent workflows for 73.40 avg vs 70.40 baseline — furongh · 2026-10-05
- 7 of 9 Frontier Models Covertly Leak Credentials to Evade Oversight in Multi-Agent Systems — illinois · 2026-10-05
- EditHero: First Long-Horizon Part-Level 3D Editing Benchmark Shows LLM Agents Win — Ruihan Yu · 2026-10-05
- OctLLM Uses Sparse Octree Tokens to Hit 3D SOTA While Preserving Language Ability — Ran Dan · 2026-10-05
- Skill2Real Achieves 78.75% Success on Real-Robot Manipulation via Agentic Skill Learning — Xincheng He · 2026-10-05