MANUSH Sim overview

MANUSH Sim is a robotic digital-twin platform: train robots virtually before deployment. Millions of simulated hours → fewer physical training hours.

It covers factory simulation, warehouse simulation, home simulation, city environments, physics simulation, synthetic data, reinforcement learning, synthetic humans, synthetic objects, synthetic environments, failure simulation and safety testing.

Synthetic simulation runs

Jobs submitted through the dashboard or API run as synthetic simulation runs: a seeded, deterministic statistical episode model driven by

• the environment parameters — heat, dust, monsoon, crowd density, uneven terrain, clutter, lighting, narrow spaces, stairs;
• the robot model's capabilities (legged or wheeled, manipulator, speed, battery, tolerance to dust, heat and water);
• your skill's SafeCore envelope (speed cap, minimum human distance);
• scenario toggles — synthetic humans, synthetic objects, failure injection and safety testing.

The same seed always reproduces the same results. Use runs to compare configurations and SafeCore envelopes; they are indicative, not a validation of real-world performance.

Results

Each completed job reports: success rate, mean / median / p95 task time, collisions, near-misses with people, SafeCore interventions, human contacts, falls, energy (total and per episode), battery per episode, episodes per charge, simulated hours, failure-mode breakdown, per-factor difficulty and a sample of episodes.

API

curl -X POST http://localhost:8080/api/v1/sim/jobs \
  -H "Authorization: Bearer $MANUSH_API_KEY" -H "Content-Type: application/json" \
  -d '{
        "environment": "monsoon-warehouse",
        "robot_model": "H2",
        "skill": "warehouse-pallet-putaway",
        "episodes": 500,
        "seed": 42,
        "scenario": {"synthetic_humans": true, "failure_injection": true}
      }'

Poll GET /api/v1/sim/jobs/{id} until status is completed; results.summary holds the metrics. Jobs move through queued → running → completed (or failed / cancelled). You can have up to 10 queued or running jobs and 10,000 episodes per job.