Search papers, labs, and topics across Lattice.
Carnegie Mellon University
1
0
2
Instruction blindness in VLA models can be mitigated by optimizing for flatter loss landscapes, leading to over 60% better adherence to language instructions.
