rl_studio
Reinforcement learning workspace. Train manipulation and navigation policies with curriculum-based RL — reward curves, ablations, and checkpoints without touching a training cluster.
$ isduit init --profile robotics-ai --mode production
ISDUIT is the engineering platform for Robotics AI & automation — train control policies in physics-accurate simulation, deploy to real fleets with staged rollouts, and orchestrate every mission from one console.
plugs into
$ isduit telemetry --arm arm-07 --stream live
✓ stream attached
$ ls ~/modules --verbose# the platform
Reinforcement learning workspace. Train manipulation and navigation policies with curriculum-based RL — reward curves, ablations, and checkpoints without touching a training cluster.
GPU-accelerated physics simulation. Validate 10,000 edge cases per hour before anything touches hardware.
Route missions across robots, cells, and sites with constraint-aware scheduling.
ISO 10218 / TS 15066 compliant speed-and-separation monitoring, enforced at the controller level.
Joint torques, force readings, and vision streams at 1 kHz — stored, replayable, diffable.
Python, ROS 2, and REST bindings for everything on the platform. Type-safe clients generated from your fleet's schema.
$ isduit workflow --show# the loop
$ isduit sim spawn --cell line-4 --replicas 256Spin up parallel digital twins of your cell. Policies train against real physics — friction, backlash, and sensor noise included.
$ isduit deploy rollout --canary 5%Staged canary releases to hardware. Roll back to any checkpoint in seconds if a metric regresses.
$ isduit watch --site all --severity warnWatch every axis, every mission, every site from one console. Anomalies surface before they become stoppages.
── each stage is a git-backed checkpoint; every policy version is reproducible end-to-end
$ isduit quickstart --urdf ./cell.urdf
Import a URDF, pick a pretrained policy, and run your first simulated mission in under an hour.