Junli Chen

The Speedup Paradox: Rethinking Inference Speed-Quality Trade-off in Embodied Tasks

Yujin Wang, Junli Chen, Yixuan Li, Shunan Dong, Huazhong Yang, Yongpan Liu, et al.

Embodied foundation models have recently been widely used to improve robot generalization and task success rates. Previous works apply lossy efficient-inference techniques such as quantization, pruning, and asynchronous inference, accepting small action quality degradation in exc…Read more