Harbin Institute of Technology

“Lead author of the paper, driving the research on real-to-sim-to-real personalization for VLA models.”

“Splat2Act: From Captured Scenes to Personalized VLA Manipulation”

“Senior author and faculty member. His lab is actively working on EmbodiedSplat and related real-to-sim navigation and manipulation, making him a key figure in Gaussian Splatting for robotics.”

“GT-VLA: Target-Conditioned Trace Guidance for Generalizable Robotic Manipulation”

“Sriram Vishwanath... Senior author and PI. Vishwanath is a well-known researcher in information theory, edge computing, and AI systems.”

“Jing-Chen Peng... Co-first author, equal contribution. Co-designed the trace-conditioned action learning and data augmentation strategies.”

“translates high-level VLM guidance into visual traces to steer low-level VLA execution on unseen tasks and configurations”

“Co-author and professor at Georgia Tech. His lab focuses on legged robot learning, motion retargeting, and physics-based character control.”

“(Listed as the corresponding author and senior researcher overseeing the project.)”

“Model-based approaches tackle granular terrain through terradynamics, most notably resistive force theory (RFT) [11]-[13]...”

“"Several recent systems iteratively co-improve a policy with a world model [38, 39, 40, 24]. All of these methods update the policy parameters... Q-Planning departs from this consensus."”

“On real bimanual robots, this loop improved success rates from 40% to 90% (stack-cups) and 25% to 80% (insert-wallet) in just five iterations”

“Principal Investigators (PIs): Jianfei Yang and Jiachen Li. The PIs provided project supervision and research direction”

“Jie Liu | Harbin Institute of Technology | Co-corresponding author | jieliu@hit.edu.cn; co-leads the project from the HIT side”

“Corresponding author email: shaorui@hit.edu.cn (Paper header). Referenced in [5], [6], [7], [40] within the same paper.”

“Instead of predying what the future looks like, ΔVLA predicts how the world changes — and that shift in framing delivers state-of-the-art manipulation performance at 3x faster training speed than comparable approaches.”