AffordanceWAM: Affordance-Aware Joint World-Action Modeling for Robot Manipulation
We introduce AffordanceWAM, an affordance-aware generative World Action Model that represents object-centric spatiotemporal affordance through Scalar Affordance and Affordance Heatmap, within the generated future World.
ProofPaper ↗
Key points
- Generalizable robot manipulation requires predicting how a scene will evolve, identifying where interactions are feasible, and determining how to act.
- Built on a pretrained video diffusion Transformer, AffordanceWAM uses separately parameterized World and Action Experts, coupled through Masked Joint Self-Attention, to jointly predict future RGB observations, Scalar Affordance fields, Affordance Heatmaps, and continuous robot actions under a unified flow-matching objective.
- Under fixed robot supervision, RoboCasa performance improves monotonically as affordance-annotated human video scales.
- These results support affordance as an effective interface for both vision-language-action learning and human-to-robot transfer.
Sources (1)
- [1]AffordanceWAM: Affordance-Aware Joint World-Action Modeling for Robot ManipulationHugging Face Daily Papers · Oct 18, 12:00 AM
We introduce AffordanceWAM, an affordance-aware generative World Action Model that represents object-centric spatiotemporal affordance through Scalar Affordance and Affordance Heatmap, within the generated future World.
Generalizable robot manipulation requires predicting how a scene will evolve, identifying where interactions are feasible, and determining how to act.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-MXFP8
- Oct 16, 2026NestedVLA: Learning to Consolidate and Generate Skills for Vision-Language-Action Models
- Oct 16, 2026MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models
- Oct 15, 2026When Plans Change, Remember: Replanning-Guided Visual Memory for Vision-Language-Action Models
- Oct 15, 2026Designing Transformers for Continuous Diffusion Language Models
- Oct 15, 2026What Should Robot Policies Learn to Predict? Rethinking Supervision Targets in Vision-Language-Action Learning