Heterogeneous Compute
"Never trust a computer you can't throw out a window." - Steve Wozniak
Vision Language Action: Model Development & Validation
version 2.9Vision-Language-Action explores how modern VLA models can translate perception, language instructions, and robot state into real physical behavior. Part 1 focuses on building and validating this pipeline using SmolVLA, the LeRobot framework, and a pair of SO-101 3D-printed robotic arms. This phase covers data acquisition, dataset quality, model fine-tuning, inference, real-time control, and systematic validation of a pick-and-place task. The goal is to understand not only whether a VLA can successfully control a physical robot, but also why it succeeds or fails and how factors such as training data, starting position, inference latency, camera input, and control software affect the resulting behavior.
Learn MoreCore Concepts
Natural language + vision models controlling real robotic arms for precise manipulation.
Vision Language Action: Inference at the Edge
version 2.9Part 2 takes the trained SmolVLA model and moves the research from development and validation into real-world edge deployment. Using custom inference and control software, the same robotic application is evaluated across heterogeneous compute platforms, including Intel Core Ultra 3 . The focus shifts to understanding how inference workloads map onto different CPUs, GPUs, and NPUs, how execution characteristics vary, and what is required to reliably deploy a physical robotic application outside of the original training environment. The goal is to determine not simply whether the model can run, but how effectively the complete system can be deployed across different classes of edge hardware.
Learn MoreCore Concepts
Natural language + vision models controlling real robotic arms for precise manipulation.
Precision Engineering
High-reliability embedded systems, FPGA design, and robotics platforms built with uncompromising attention to detail.
Verified Performance
Benchmarked, stress‑tested, and validated across real‑world robotics and compute workloads.
Robotics & Control
Physical systems, precise motion, and intelligent actuation.