SenseTime has open-sourced the SenseNova-RoboRSI and 27B robotic models, while testing has revealed cost bottlenecks
SenseTime open-sourced two projects on Oct. 21: the embodied AI evaluation framework SenseNova-RoboRSI and the 27B-parameter embodied model SenseNova-RoboRSI-27B-Preview. Testing found that, after providing access to training demonstrations and increasing the step limit to 10 times the official setting, inference costs were extremely high. GPT-6-Astra Tested on 12 Simulated Benchmarks and 8 Physical Robots According to SenseTime’s test report, when access to training demonstrations was provided
SENSETIME+3.73%
Bisc·10-21 02:53
