Publikation
Kintsugi: Learning Policies by Repairing Executable Knowledge Bases
Teng Cao; Yu Deng; Hikaru Shindo; Quentin Delfosse; Lanxi Wen; Suli Wang; Jannis Blüml; Christopher Tauchmann; Kristian Kersting
In: Computing Research Repository eprint Journal (CoRR), Vol. abs/2605.09487, Pages 1-35, arXiv, 2026.
Zusammenfassung
Modern embodied agents achieve impressive performance, but their task knowl-
edge is often stored in neural weights, latent state, or prompt-bound memory,
making individual policy knowledge difficult to inspect, validate, recombine, and
reuse. We introduce Kintsugi, a white-box policy-learning framework that treats
embodied policy improvement as verifier-gated construction of a typed executable
Knowledge Base (KB). Kintsugi represents task-level policy knowledge as com-
posable typed entries—predicates, operators, policy schemas, monitors, recovery
rules, experience records, and goals—and improves this artifact through local-
ized typed edits induced from rollout evidence, rather than relying on test-time
language-model reasoning. Between rollouts, a tool-constrained agentic editing
loop diagnoses trajectory failures, localizes them to editable KB layers, and pro-
poses candidate edits. A deterministic verification gate admits an edit only when
the candidate type-checks, the resulting KB executes, and focused validation suc-
cess or trajectory-health metrics improve without violating protected-regression
checks. At inference, the accepted KB is executed by a deterministic symbolic
executor with zero LLM calls. Across long-horizon text-agent benchmarks and rep-
resentative object-centric manipulation settings, Kintsugi achieves strong endpoint
performance while preserving inspectability, local editability, and verifier-gated
deployment. These results suggest that embodied policy improvement can be
organized around executable task knowledge.