AI
REALM: A Coarse-to-Fine Generative Framework for Embodied Reactive Listening
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 26/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 04:23:25 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 04:18:24 UTC.
Briefings
The REALM framework addresses embodied conversational AI challenges by combining history-aware temporal alignment with stochastic expression refinement. It introduces a Reactive Gated Speaker-Listener Fusion module that fuses listener motion history with speaker audio via delay-centered attention and adaptive gating. A coarse decoder predicts base motion trajectories augmented by audio-conditioned stochastic residuals in the expression subspace while retaining pose parameters. Evaluated on ViCo and L2L datasets, the model shows improvements over baselines across motion-quality metrics. Deployment on an Ameca humanoid robot and a perceptual user study confirm applicability to physical embodiment.
Key points
- The REALM framework combines history-aware temporal alignment with stochastic expression refinement to address embodied conversational AI challenges.
- A Reactive Gated Speaker-Listener Fusion module uses delay-centered attention and adaptive gating to fuse listener motion history with speaker audio.
Context and limitations
Evaluation results are limited to ViCo and L2L datasets; deployment on an Ameca humanoid robot and perceptual user study results are not yet fully quantified.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 04:23:25 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifclkzf4tcky4fatvkhcd7oeainmw6hg4cpv7uuelaoqsfd5zqgji