10/05/2026
William & Mary students and faculty wrapped up an exciting week at in Pittsburgh! Our team presented papers and posters spanning vision-language models, world models, autonomous driving, and immersive robotic telepresence.
💡 Fast Enough to Act: Spatio-Temporal Visual Token Merging for Low-Latency Robotic VLMs and VLAs
Junzhou Chen, Jindong Wang, Gang Zhou
For robots to act in the real world, VLMs and VLAs must be fast as well as smart. Our ST-Merge work cuts spatio-temporal visual token redundancy to make robotic VLM/VLA inference markedly lower-latency.
💡 Noise-Informed Conformal Prediction for Trajectory World Models
Fujie Zhang, Yandong Luo, Yuchen Wang, Chuchu Chen, Lu Gan, Huajie Shao
Trajectory data collected from robotic systems are inevitably corrupted by unbounded and temporally correlated observation noise. We propose Noise-Informed Conformal Prediction (NICP) to provide calibrated uncertainty estimates for trajectory world models under such noisy observations.
💡 STADA: Specification-based Testing for Autonomous Driving Agents
Joy Saha, Trey Woodlief, Sebastian Elbaum, Matthew B. Dwyer
This work presents a test generation method based on formal specifications of driving rules. We demonstrate that STADA is capable of generating tests that cover twice as many specification-relevant scenarios than prior methods and do so much more efficiently.
💡 Toward Real-time High-fidelity Immersive Robotic Telepresence via Online 3D Gaussian Splatting
Jingxin Du, Mo Xu, Yaxin Hu, Xuebin Sun, Bilge Mutlu, Kevin Ponto
This work presents a novel online 3D Gaussian Splatting (3DGS)-based virtual reality (VR) robotic telepresence system that effectively balances visual quality, system performance, and display immersiveness through a careful co-design spanning the system’s algorithms, architecture, and pipelines.
Congratulations to everyone! Looking forward to seeing what the next year brings.
