RT DW A1 Wang, Tianyu. T1 Learning Generalizable Robot Policies by Understanding Semantics and Logic from Task Demonstrations