Search papers, labs, and topics across Lattice.
State Key Laboratory of CAD&CG, Zhejiang University
8
0
12
1
Achieving up to 100% success on complex tasks without the need for search, INTACT redefines how we approach intent-to-action learning in dynamic environments.
As AI systems become more capable and opaque, the role of symbolic methods shifts from computation to essential interfaces for human oversight and trust.
Alignment isn't enough: truly safe AI demands robust runtime controllability, which current methods often fail to provide.
Image editing, surprisingly, holds the key to robots that can nimbly manipulate objects in the real world.
A single malicious message can trigger a self-replicating worm, ClawWorm, that autonomously infects and propagates across entire LLM agent ecosystems, even surviving agent restarts.
Give your memory-less VLA policy a brain: TempoFit retrofits temporal context by cleverly reusing existing attention keys and values, boosting long-horizon task success without retraining or adding latency.
Backdoors aren't just for attacks anymore: B4G shows how they can be flipped to enhance LLM safety, controllability, and accountability.
Achieve a 37% success rate boost in precision insertion by intelligently fusing human corrections with learned policies, triggered only when force discrepancies signal impending failure.
