Robotic Foundation Models
The π0 foundation model continues to evolve, demonstrating enhanced generalist capabilities in physical manipulation tasks. The latest iteration, π*0.6, is trained using Recap (RL with Experience & Corrections via Advantage-conditioned Policies), which combines demonstrations, expert corrections, and autonomous experience. This approach significantly improves robustness and throughput on complex real-world tasks like making espresso drinks, assembling boxes, and folding laundry, enabling practic. The FAST tokenizer has been developed to compress action sequences using the discrete cosine transform (DCT) combined with byte pair encoding (BPE), allowing for more efficient training of autoregressive transformer policies for dexterous robotic tasks. This results in a 10x compression over prior action tokenization approaches and enables training 5x faster than previous models, with policies generalizing to new environments and instructions zero-shot.
Timeline
