AI

Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 04:03:24 UTC.

Briefings

A new self-supervised method called Keyframe Mnemonics discovers critical observations to enable behavior cloning in non-Markovian environments. The approach learns an objective from past observations to select keyframes, allowing policies to retain context over infinite horizons while using minimal working memory. Evaluation on synthetic domains shows 100% success rates, and on a real robot manipulation benchmark, it improved average success rates by 13.9% compared to the strongest baseline across 23 tasks.

Key points

• Keyframe Mnemonics discovers information-critical observations by learning an objective from randomly sampled past observations to guide keyframe selection.
• The method achieves 100% success rates on synthetic memory domains and improves average success rates by 13.9% on a real robot manipulation benchmark across 23 tasks.

Context and limitations

Evaluation on synthetic memory domains and a real robot manipulation benchmark; performance on other horizons or environments remains untested.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/10/2026, 04:40:59 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiejr357cyzwnkbap3tnbpfyicsaoytfmnrrutq5tktznhegx3d7rm