VoxelDyn Labs is founded and led by Sam Harrison, uniting deep academic rigor with hands-on GPU kernel engineering to build physical AI foundations.
Sam Harrison is a computer vision and robotics researcher formerly at Stanford AI Lab with over 8 years of specialized experience in real-time CUDA acceleration, 4D Gaussian dynamics, and physical simulation engines. Sam has authored 6+ CVPR and NeurIPS publications on neural radiance and continuous spatial kinematics, accumulating over 1,800 academic citations.
Recognizing that 2D video models cannot satisfy the low-latency and physics constraints of physical robotics, Sam founded VoxelDyn Labs as a Delaware C-Corp to develop native 4D spatio-temporal foundation models powered by high-throughput GPU silicon.
"The internet is filled with language and 2D images, and current AI models have mastered them. But robots do not live on the internet. Robots operate in an uncompromising physical world governed by Newton's laws, friction, inertia, and 3D space."
"Trying to run an autonomous robot using a 2D diffusion model is like asking a human driver to navigate an obstacle course while only seeing disconnected still photographs sent with a two-second delay. It cannot react to sudden changes, and it cannot guarantee safety."
"At VoxelDyn Labs, we are building the continuous, native 4D foundation. By fusing 4D Gaussian representations directly with GPU acceleration and TensorRT, we are giving machines the reflexive spatial intuition they need to interact safely with the physical universe."
VoxelDyn Labs Inc. is a registered Delaware C-Corporation headquartered in San Francisco, California, purpose-built for venture-backed deep-tech growth.
Actively scaling high-throughput cluster resources to accelerate 14B and 32B parameter foundation model pre-training across H100 GPU nodes.
All physical world model predictions feature deterministic geometric bounding envelopes to guarantee collision-free execution on real robot hardware.