Acceptance Testing · Robots · Policies · AMR Fleets
FAT/SAT for Robots.
Factory and Site Acceptance Testing for the AI policies and AMR fleets that run on top of your hardware. The same decades-old rigor procurement gives the gripper — applied to the software and the fleet behavior the hardware runs.
The gap
Hardware gets formal acceptance testing. The AI running on it doesn't.
What we test on the robot today
- ▪Repeatability ± tolerance
- ▪Payload @ reach
- ▪Cycle time vs datasheet
- ▪ISO 10218 / TS 15066 safety
- ▪Power, comms, E-stop wiring
— decades old. Standard procurement practice.
What gets tested on the policy + the fleet
- ▪Task success rate at scale
- ▪Trajectory smoothness + jerk
- ▪Force compliance under contact
- ▪AMR throughput vs WMS spec
- ▪Multi-agent collision + deadlock
- ▪Dock-and-charge reliability
— today: "looks good, ship it." We close that gap.
Two product lines under FAT/SAT
Simulation-driven acceptance testing for both the policy and the fleet.
01 · Policy FAT/SAT
For AI manipulation, grasping, and contact-rich workflows.
Your manipulation policy runs in our containerized test harness against randomized scenes in Isaac Sim, MuJoCo, or Gazebo. Deterministic seeds. N trials per rubric dimension. Per-trial video replay.
Pilot hardware: UR5e · Robotiq grippers · ATI F/T sensors
02 · AMR FAT/SAT (Digital Twin)
For AMR fleets, throughput, and multi-agent coordination.
We build a digital twin of your facility — racking, dock positions, charger layout, WMS integration — and validate fleet throughput, multi-agent collision behavior, and dock-and-charge cycles before a single bot rolls onto your floor.
Pilot fleet support: MiR · Locus · Geek+ · OTTO · roadmap: Fanuc, KUKA
The FAT/SAT report
Per-trial scoring against an acceptance rubric.
A real test report, not a marketing deck. Pass criteria match what procurement writes into a robotics RFP. Below is a representative Policy FAT rubric — exact thresholds get scoped to your workload.
| Dimension | Pass criterion |
|---|---|
| Task success rate | ≥ 95% over N=50 trials |
| Trajectory smoothness | avg jerk < 5 m/s³ |
| Cycle time | ≤ 8 s p95 |
| Path efficiency | ≤ 1.2× geodesic distance |
| Force compliance | peak ≤ 20 N · sustained < 1 s |
| Off-limit contacts | 0 |
Representative rubric. AMR FAT/SAT swaps in throughput, dock cycles, collisions, and WMS-integration metrics.
How it works
Three phases. Simulation first. Floor second.
01
FAT
Your policy or digital-twin model runs in our containerized harness on cloud GPUs against randomized scenes. Deterministic seeds. N trials per rubric dimension.
02
Iterate
You get structured reports (YAML + PDF), per-trial video replays, and a regression history across runs. Fix the failure modes. Re-test in minutes, not weeks.
03
SAT
When FAT passes, we re-run the same harness on your physical workcell or warehouse floor for site acceptance. Customer signs off the rubric. PO closes.
Methodology
Real measurements. Reproducible runs. Sim-to-real discipline.
Test rigs use force-torque sensor monitoring for compliance dimensions, off-limit contact detection for safety dimensions, and domain randomization across three simulators (Isaac Sim, MuJoCo, Gazebo) for sim-to-real robustness.
Everything runs on ROS 2 with deterministic seed reproducibility — re-run the same seed, get the same trial result. AMR digital twins integrate through each brand's published fleet API and (where available) OPC-UA or WMS integration layers so the twin exercises the same control surfaces the live fleet would.
Who it's for
Three buyers, one rubric.
Integrators
Selling AI-driven cells.
Pass FAT/SAT, win the PO. Quantitative sign-off your customer can put in a procurement file.
End customers
Writing policy acceptance into the RFP.
Stop accepting "the demo worked." Write quantitative pass criteria into the contract and hold vendors to them.
Robot OEMs
Third-party policy validation.
Independent rubric run by a third party. Use the report in sales conversations with skeptical procurement teams.
Run a FAT pilot on your policy or fleet.
One workload. One rubric. One report. No strings. Typical pilot scopes in 2–4 weeks and gives you a procurement-grade artifact at the end — pass or fail.