01 / DEVELOPER WORKBENCH & SIM-3
Explore the newly qualified SIM-3 3D Kinematic Tree Studio, inspect analytical equations of motion in WebAssembly, build the $150 Fast Spine microcontroller, and verify physical AI execution.
An interactive, zero-dependency 3D browser workbench powered by our analytical C++ dynamics engine compiled to WebAssembly (ABI 2). Inspect multi-link articulated chains with live matrix readouts.
seedcore.mini-sim.tree-model.v1 with canonical SHA-256 digests and JSON exportPRACTICAL HARDWARE ARCHITECTURE
Mini-robots do not need $3,000 industrial SoCs. Build an agile, safe-by-physics physical AI embodiment with off-the-shelf maker electronics.
Microcontroller: ESP32-S3 or Raspberry Pi RP2040 ($6).
Actuators: 4–6x Feetech STS3215 magnetic encoder serial bus servos ($60–$80).
Power: 2S LiPo (7.4V) or 5V USB-C PD ($15).
Chassis: 3D-printed PETG / PLA / TPU ($10).
Camera: USB UVC webcam or ESP32-CAM ($12).
Total BOM: ~$150. Accessible for makers, students, and children.
The edge MCU fast spine runs a deterministic hardware timer interrupt. It receives receding-horizon action chunks (30–50 waypoints) from the host brain over UART/WebSerial.
Evaluates smooth cubic Hermite spline trajectories locally and drives servos over 1 Mbps half-duplex UART bus with zero communication jitter.
Includes hardware watchdog: brings motors to rest if packets cease for > 100ms.
Using the W3C WebSerial API, any browser running on a Chromebook, Mac, or Windows PC can connect directly to the robot's USB-CDC port without installing drivers, ROS 2, or Linux.
Deploy trained neural policies, send action chunks, inspect live proprioception, and verify cryptographic receipts directly in the web browser.
Read the Developer Blueprint ↗BEGINNER & MAKER QUICKSTART
Building your first desktop robot shouldn't require electrical engineering. Here is exactly how the consumer hardware connects on your desk.
Each Feetech STS3215 bus servo features two identical 3-pin snap sockets. Plug the first cable from your ESP32 board to Servo 1, then daisy-chain from Servo 1 to Servo 2, and Servo 2 to Servo 3. Total wiring time: under 3 minutes.
Plug the ESP32 into your computer via USB-C. Open 3D Tree Studio in Chrome, Edge, or Opera. Click "Connect USB" and pick your device in the standard browser popup. No drivers or baud rate configuration needed.
Move the joint sliders on your screen—the physical servos track each angle in real time. Switch on "Teach Mode" to gently guide the arm by hand; the browser records your exact motion path for automated playback.
PHYSICAL AI REFERENCE FOUNDATION
Bridging simulated digital twins to physical mini-robot embodiments.
Versioned MJCF package with SHA-256 manifest: 38 STL meshes, 14 actuators, total mass 0.737 kg. Validated against upstream Onshape CAD geometry.
Automated tests verify body tree hierarchy, joint limits, hinge ranges, and forward kinematics reproducibility.
Inspect Model Package ↗Draft, run, and compare bounded simulations natively. Each experiment records joint state, simulated sensors, contacts, and model/recipe cryptographic identities.
Provides agentic tools for autonomous policy inspection and parameter tuning before hardware deployment.
Run Native Workbench ↗Current active engineering milestone: formalize the 61-dim observation vector (48 state/history + 13 commands), 14-dim action vector, home poses, and zero-drift startup verification.
Directly paves the way for M2 (governed velocity tokens) and M3 (action-bound audit receipts).
Follow Integration Plan ↗SIMULATOR EVOLUTION ROADMAP
Each milestone is gated by deterministic mathematical invariants and reproducible test fixtures.
WebWorker isolation, transferable ArrayBuffers, immutable model digests, and reproducible lesson traces.
Zero-allocation C++ two-link dynamics kernel, dual native/Wasm build, JS/native trace equivalence (< 1e-12 error), and resumable checkpoints.
1-, 2-, and 3-link serial chains, Cramer cofactor matrix inversion, skew-symmetric Coriolis vector, 3D perspective projection with orbit controls, live M(q)/J(q) inspectors, 78/78 tests passing.
Qualify rigid contacts, friction cones, joint stops, floating base dynamics, and simulated IMU sensors.
Headless multi-instance parallel stepping, memory bounds, and Gymnasium/Python reinforcement learning wrappers.
Full integration with SeedCore capability tokens, denial cases, and closed-loop audit receipts.
Source: SeedCore Simulator Implementation Plan. Verified across Node.js, CMake/CTest, and Python pytest.
TAKE THE NEXT STEP
No installation, no login, no remote dependencies. Open the lab and explore nonlinear multi-link articulated dynamics directly in your browser.