AIReal-Time EXPO-FT targets latency with a small, real-time edit policy
The team says its reinforcement-learning approach outperforms RTC by a significant margin across the board, pairing a base model with a real-time edit policy.
TLDR
The team behind Real-Time EXPO-FT says its approach builds on EXPO-FT: a base vision-language-action model operates on older images while a small edit policy runs in real time. The project focuses on reinforcement learning for dynamic real-world manipulation tasks. The team claims it outperforms RTC by a significant margin across the board.
Combined views
9K
1 Source, first seen 5h ago
92 likes
Real-Time EXPO-FT targets latency with a small, real-time edit policy
The team says its reinforcement-learning approach outperforms RTC by a significant margin across the board, pairing a base model with a real-time edit policy.