Search papers, labs, and topics across Lattice.
This paper introduces Tycho, a coding-agent system that enhances skill acquisition in unfamiliar games by constructing game-specific models through interactive problem-solving. By formalizing game environments as parameterized rendered deterministic Moore machines, Tycho enables agents to efficiently separate actionable observations from non-essential frames, leading to improved action efficiency. The results show that the actor-requested delegation policy achieves a mean Relative Human Action Efficiency (RHAE) of 88.49, with top-performing models reaching perfect RHAE scores while using significantly fewer actions than human baselines.
Tycho's innovative approach to active abstraction allows agents to achieve perfect action efficiency while navigating complex game environments, outperforming human players significantly.
ARC-AGI-3 turns abstraction into an interactive problem of skill acquisition. A player must infer an unfamiliar game's rules, hidden state, and goal while maintaining action efficiency because every move counts. We formalize these environments as parameterized rendered deterministic Moore machines and introduce Tycho, a coding-agent system that constructs and uses game-specific models during interaction. Tycho separates actionable observations from intermediate animation, level-completion, and game-over frames. From this structured history, an agent can model, test, plan with, repair, or bypass a free-form executable hypothesis. In one matched public-set run per policy, we compare four orchestration policies on all 25 public games using Claude Opus 4.8 under matched inference budgets. Actor-requested delegation to a model builder obtains the highest observed mean Relative Human Action Efficiency (RHAE), 88.49. With this selected policy, GPT-5.6 Sol and Opus 5 both reach 100.00 RHAE and complete all 183 levels. Their game-balanced first-run human-replay midranks are 98.5 and 100.0. Opus 5 uses 61% fewer scored actions than the aggregate official human baselines. Automatic repair after verification failures produces models that reproduce observed transitions much more accurately, yet reaches only 83.07 RHAE. Transition match indicates whether a simulator reproduces observed dynamics, not whether it has identified the objective or improves the next action. Strong play also requires deciding when to construct, repair, use, or bypass a model. We call this joint problem active abstraction: generating a testable model from costly interaction and deciding when acquiring or using it is worth its cost.
Amazon Science