Robotics and research software engineering — embodied systems, simulation, and the data infrastructure underneath them.
Previously RSE at Stanford (METER-AI, Doerr School of Sustainability), architecting cloud pipelines over 14M+ records and leading the open-sourcing of a methane emissions dataset now used by Planet and CarbonMapper. Currently focused on physical AI: the NVIDIA Isaac and Omniverse stack, vision-language-action models, and sim-to-real transfer — and working toward the NVIDIA OpenUSD Development Professional certification.
Exploring Gemini Robotics-ER (Google DeepMind) for embodied reasoning — spatial grounding, and how a general vision-language model's scene understanding maps onto a real robot's action space.
Most interested in: AI apps and agents · fine-tuning and post-training workflows · research, benchmarking, and experimentation · robotics and physical AI · MCP and agent tooling protocols · data infrastructure at scale · simulation and synthetic data · graph-based systems
| Chip's Robot Adventure |  |
Browser maze game synchronized in real time to a physical LEGO MINDSTORMS Inventor hub over Bluetooth |
The mgraph.* suite — a modern ES module refactoring of the ngraph.* ecosystem, rebuilt with
typed entry points, Vitest, and Rollup bundles. All eight packages are published to npm.
| mgraph.forcelayout |  |
Force directed graph drawing layout |
| mgraph.graph |  |
Modern graph data structure for JavaScript |
| mgraph.random |  |
Seeded random number operations for mgraph.* |
| mgraph.events |  |
Modern eventing library for Node.js and browsers |
| mgraph.merge |  |
Simple merge utility to extend objects without extra dependencies |
| mgraph.fromdot |  |
Load DOT files into mgraph.graph |
| mgraph.generators |  |
Graph generators library |
| mgraph.fromjson |  |
Load graphs from a simple JSON format |
| corrigible-mcp-tools | |
Decorators adding confidence scoring, circuit breakers, provenance logging, and consequence-based checkpoints to MCP tools |
| cerebras-neurips-2024-chatbot |  |
Conference chatbot built on Cerebras inference, deployed for NeurIPS 2024 |
Not my code — the stack I read, run, and build against, grouped by where it sits in the pipeline.
Entries marked * are ones I've forked and am actively running.
| cosmos |  |
World foundation models, datasets, and tooling for physical AI |
| Isaac-GR00T |  |
Foundation model for generalist humanoid robots |
| IsaacLab |  |
Unified robot learning framework built on Isaac Sim |
| IsaacSim |  |
Robotics simulation application built on Omniverse |
| warp |  |
Python framework for GPU-accelerated simulation and robotics |
| curobo |  |
CUDA-accelerated motion generation and collision-free planning |
| RoboLab |  |
Simulation benchmark for evaluating generalist robot policies |
| physicsnemo |  |
Physics-ML framework for training surrogate and operator models |
Where the models meet the robot — hardware-accelerated ROS 2 perception on the edge.
| isaac_ros_visual_slam |  |
Visual SLAM and odometry built on GPU-accelerated cuVSLAM |
| isaac_ros_nvblox |  |
Accelerated 3D scene reconstruction feeding Nav2 local costmaps |
| isaac_ros_pose_estimation |  |
Deep-learned 3D object pose estimation |
| isaac_ros_cumotion |  |
Accelerated arm motion planning and control |
| isaac_ros_nitros |  |
Zero-copy transport for hardware-accelerated ROS 2 graphs |
| jetson-containers |  |
ML container builds for Jetson and JetPack-L4T |
| TensorRT |  |
High-performance inference SDK — the deployment target for edge policies |
| holoscan-sdk |  |
Low-latency sensor processing SDK for streaming workflows |
Pursuing the NVIDIA OpenUSD Development Professional certification — scene description as the interchange layer between simulation, synthetic data, and robot learning.
| OpenUSD |  |
Universal Scene Description — the reference implementation |
| LearnOpenUSD |  |
Tutorials and reference material for OpenUSD |
| OpenUSD-Code-Samples |  |
Common code snippets for OpenUSD |
| kit-app-template |  |
Omniverse Kit application scaffold |
| PhysX |  |
NVIDIA PhysX SDK |
| lerobot* |  |
End-to-end learning for real robots, with datasets and pretrained policies |
| openpi |  |
Open vision-language-action models from Physical Intelligence |
| real-time-chunking-kinetix |  |
Real-time execution of action-chunking flow policies |
| aloha |  |
Bimanual teleoperation hardware and software stack |
| dm_robotics |  |
Libraries, tools, and tasks used at DeepMind Robotics |
| dreamerv3* |  |
Mastering diverse domains through world models |
| mujoco |  |
Multi-joint dynamics with contact — general purpose physics simulator |
| mujoco_playground |  |
GPU-accelerated robot learning and sim-to-real transfer |
| mujoco_warp |  |
GPU-optimized MuJoCo built for NVIDIA hardware |
| mujoco_mpc |  |
Real-time behavior synthesis via predictive control |
| Genesis |  |
Simulation platform for general-purpose robotics and embodied AI |
| drake |  |
Model-based design and verification for robotics |
| ros2 |  |
The Robot Operating System — middleware, build system, and core stack |
| navigation2 |  |
ROS 2 navigation: planning, control, recovery behaviors |
| moveit2 |  |
Motion planning and manipulation for ROS 2 |
| ros2_rust* |  |
Rust bindings for ROS 2 |
| FoundationPose |  |
Unified 6D pose estimation and tracking of novel objects |
| FoundationStereo* |  |
Zero-shot stereo matching for depth perception |
| BundleSDF |  |
Neural 6-DoF tracking and 3D reconstruction of unknown objects |
| instant-ngp |  |
Instant neural graphics primitives for fast scene reconstruction |
| ORB_SLAM3 |  |
Visual, visual-inertial, and multi-map SLAM |
| Open3D |  |
Modern library for 3D data processing |
People whose code, teaching, and taste shape how I work.
| karpathy |  |
Andrej Karpathy — nanoGPT, micrograd; teaching by building from scratch |
| lucidrains |  |
Phil Wang — hundreds of faithful paper implementations, attention in every form |
| ggerganov |  |
Georgi Gerganov — llama.cpp and whisper.cpp; inference at the edge, where robots live |
| anvaka |  |
Andrei Kashcha — creator of the ngraph.* ecosystem my mgraph.* suite modernizes |
| dusty-nv |  |
Dustin Franklin — the Jetson edge-AI container ecosystem |
| AtsushiSakai |  |
Atsushi Sakai — PythonRobotics, the canonical readable robotics algorithms |
| danijar |  |
Danijar Hafner — world models and Dreamer; this page's format borrows from his profile |
| RussTedrake |  |
Russ Tedrake — Drake and Underactuated Robotics; rigor in robot dynamics |
| mberman84 |  |
Matthew Berman — hands-on AI tooling and model evaluations |
Off GitHub: Ilya Sutskever, whose research program and talks shape how I think about scaling and generalization.
Certifications — NVIDIA OpenUSD Development Professional (in progress) · NVIDIA Agentic AI (February 2026) · Google Cloud Professional Data Engineer (December 2025) · Neo4j Certified Professional and Graph Data Science (2025) · Google Cloud Professional ML Engineer (2020) · Google Cloud Professional Cloud Architect (2018, 2020)
Powered by coffee and an unreasonable determination to extract every last token of value from an LLM. (AI-typed, human-inspired.)