Studio · AI · Reserve

The platform for the robotics economy.

Simulate and train humanoids in the browser, work with a robotics-native AI — and follow Xynoid Reserve, our on-chain reserve of robotics and physical-AI exposure.

H1
Studio robot
19 DOF
Physics sim
— $BOT
Reserve →
Unitree H1 humanoid robot rendered as a halftone illustration
01 — Xynoid Reserve

An on-chain reserve for the robotics economy.

Beyond software, Xynoid is building a long-term treasury of robotics and physical-AI exposure — starting with RoboStrategy and expanding across humanoids, automation and infrastructure, one position at a time.

HELD
Owned by the reserve
PLANNED
Intended, not yet acquired
TARGET
Potential future position
WATCHLIST
Monitored, no plan
Reserve registerOn-chain wallet
First position
— $BOT
RoboStrategy · Solana reserve wallet
AssetSymbolStatusQty
RoboStrategyBOTHELD—
Figure AI—TARGET—
Unitree—TARGET—

Targets are not holdings. Listing a company does not mean Xynoid owns its shares. No returns are promised.

02 — Xynoid Studio

Cloud robotics simulation, rendered in your browser.

A single, physically accurate environment built around the Unitree H1 humanoid — plus the training infrastructure to teach it: parallel environments, domain randomisation and GPU reinforcement learning.

Open Studio
RobotUnitree H1
DOF19
PhysicsRapier rigid-body
RuntimeBrowser · WASM
FormatOpenUSD / URDF
MiddlewareROS 2 Humble
h1_training.usdLive
XYZ 0.00 0.00 1.02
policy · walk_v42
$browser gait search · GPU training under development
Xynoid AI · h1-training
Why is H1 falling forward after 300 steps?
Xynoid

Your torso pitch reward is saturating. The ankle torque limit is clipping before the policy can recover. Three fixes:

—Raise ankle_pitch effort 32 → 48 Nm in h1.urdf
—Add pitch-rate penalty, weight -0.35
—Randomise ground friction 0.6–1.1 across 4096 envs
reward.py
rew  = 1.2 * fwd_vel
rew -= 0.35 * abs(pitch_rate)
rew -= 0.02 * torque_l2
rew += 0.5 * feet_air_time

Applied to the live sim. Restart training with the new reward?

03 — Xynoid AI

A private model that thinks in robotics.

Trained on ROS 2, MoveIt, Isaac Sim, control theory, SLAM and embedded systems. It reads your project, edits your simulation and supervises your training runs.

Generate ROS 2 nodes
Debug ROS errors
Tune H1 locomotion policies
Write URDF & MoveIt configs
Design reward functions
Read robotics papers
Inspect training runs
Analyse camera & LiDAR data
Deploy to Jetson & edge
04 — Integration

Studio, AI and Reserve — one Xynoid.

The model reads the active scene graph, sensor layout and training metrics. Ask for a change and it edits the simulation in place — no copy-pasting configs between tools.

ROS 2OpenUSDIsaac SimMuJoCoMoveIt 2PyTorchIsaac LabTensorRTONNXDockerKubernetesJetsonROS 2OpenUSDIsaac SimMuJoCoMoveIt 2PyTorchIsaac LabTensorRTONNXDockerKubernetesJetson