AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

ReCAPA: Hierarchical Predictive Correction to Mitigate Cascading Failures

Xiyin Zeng, Yuyu Sun, Haoyang Li, Shouqiang Liu, Hao Wang

ICLR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add HPCC multi-level predictive correction (ablation)ReCAPA · Behavior (EmbodiedAgentInterface)success rate: 59.3 -> 72.2Improved
add trajectory level to action+subgoal HPCC (ablation)ReCAPA HPCC · AI2-THORsuccess rate: 0.65 -> 0.75Improved
add subgoal level (ablation)ReCAPA · Behavior (EmbodiedAgentInterface)success rate: 62.8 -> 72.2Improved
use hierarchy instead of FlatHead single decoder (ablation)ReCAPA · Behavior (EmbodiedAgentInterface)success rate: 49.3 -> 72.2Improved
add Sinkhorn and Score-field alignment (ablation)ReCAPA · Behavior (EmbodiedAgentInterface)success rate: 65.8 -> 72.2Improved
use Sinkhorn alone instead of Score-field alone (ablation)ReCAPA alignment · Behavior (EmbodiedAgentInterface)success rate: 64.4 -> 66.1Improved
replace Sinkhorn with KL divergence (ablation)ReCAPA alignment · VisualAgentBench Minecraftsuccess rate: 66.7 -> 67.0Mixed
replace HPCC with PPO flat RL (ablation)ReCAPA · AI2-THORsuccess rate: 0.75 -> 0.59Hurt
replace HPCC with HIRO fixed-interval subgoals (ablation)ReCAPA · VirtualHomesuccess rate: 70.5 -> 62.7Hurt
joint training of all levels instead of separate bottom-up training (ablation)ReCAPA · AI2-THORsuccess rate: 0.59 -> 0.75Improved
train trajectory level jointly instead of freezing it (ablation)ReCAPA · Behavior (EmbodiedAgentInterface)success rate: 68.9 -> 72.2Improved
use ReCAPA hierarchical correction instead of GPT-4VReCAPA · AI2-THORcoverage: 0.97 -> 0.95Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy