Search papers, labs, and topics across Lattice.
13
0
15
0
JITOMA slashes active graph size and latency, ensuring robots can efficiently navigate long-horizon tasks without perceptual overload.
Time-derived progress labels can mislead robotic learning, but UR-VC corrects these inaccuracies to enhance task success in real-world manipulation.
DualEval reveals that unifying static and preference-based evaluations can lead to more reliable model rankings and deeper insights into item performance.
Post-training on synthesized safety-critical scenarios can dramatically enhance the reliability of autonomous driving systems, reducing failures in rare but critical events.
Iterative refinement with feedback-driven learning allows AutoDecompiler to significantly enhance the accuracy of binary decompilation, outperforming traditional single-turn models.
RoboNaldo slashes free-kick shot errors by nearly 50% while achieving ball speeds that rival professional soccer players.
On-device LLMs can now drive real-time recommendation improvements, unlocking faster adaptation to evolving user intent without cloud reliance.
Recommendation agents can achieve state-of-the-art performance by personalizing not just user memory, but also the reasoning process itself through self-evolving, user-specific policy skills.
Splitting AI systems into planning, reasoning, and execution layers on specialized hardware slashes latency and energy use by 70% compared to monolithic approaches.
Tactile-aware robot manipulation gets a serious upgrade: TAMEn's wearable interface and data pipeline more than double task success rates in complex bimanual tasks.
LLMs can achieve massive performance gains on reasoning and knowledge-intensive tasks simply by iteratively refining their answers using pseudo-labels derived from unlabeled data.
Humanoids can now play ping-pong using *only* onboard cameras, pulling off whole-body smashes and crouch shots with impressive agility.
VLMs get a 24% performance boost and run 56% faster on robot manipulation tasks by explicitly modeling action advantages and exploring multiple future paths, instead of relying on noisy foresight predictions.
