CAP: Codebook-Aligned Prediction for Tokenized Robot Policies
Overview
CAP reuses tokenizer code vectors as policy prototypes, improving task success in simulation and real robots. Tokenizer latent structure aids policy learning without changing the model architecture. arXiv:2610.09178v1 Action tokenization converts continuous actions into discrete symbols for autoregressive modeling. Conditions: Action tokenization converts continuous robot actions into discrete symbols that can be modeled autoregressively. Evidence: arXiv:2610.09178v1 Extracted Conditions: Action tokenization converts continuous robot actions into discrete symbols that can be modeled autoregressively. Quote: Action tokenization converts continuous robot actions into discrete symbols that can be modeled autoregressively. Extracted Evidence: arXiv:2610.09178v1
Generated from attributed reports · Updated 1 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
No current evidence-backed claims. Missing information remains not reported.
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsCAP: Codebook-Aligned Prediction for Tokenized Robot Policies
This research proposes CAP, a method that reuses the code vectors from a tokenizer as policy class prototypes, improving task success in both simulation and real-robot tasks. The approach maintains the tokenizer and policy backbone unchanged, showing that the latent structure of tokenizers provides valuable information for policy learning.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).