Search papers, labs, and topics across Lattice.
4
0
5
MANGO reveals that automated oracle generation can match the diagnostic power of traditional methods, transforming how we test VLA-enabled robotic systems.
Static benchmarks underestimate VLA model failures, but a new adaptive testing approach reveals up to 29.7% more critical weaknesses.
Metamorphic Testing offers a surprisingly effective, oracle-free approach to uncovering failures in vision-language-action robots, even detecting subtle issues like uncompleted tasks.
Traditional mutation analysis falls short in DL-enabled robotic software, but UAMTERS injects stochastic uncertainty to reveal critical failure modes that would otherwise go undetected.
