Skip to content

Acceptance

v1.0.0 is the day this command passes against real backends with no mocks:

git clone https://github.com/cagataycali/stobor && cd stobor
python3 -m pip install ".[all]" "mhs[nats,zenoh] @ git+https://github.com/modelhardwarestandard/python-sdk"
bash acceptance/all.sh
# script verifies backend
1 01_sim_step.py reset / step / render on the SO-101 MJCF returns a 480x640x3 frame MuJoCo
2 02_policy_checkpoint.py a genuine lerobot checkpoint answers through Policy lerobot
3 03_rollout_rtc.py a rollout with RTC replan drives the sim MuJoCo + policy
4 04_record_reopen.py a recorded dataset is finalized and re-read in a fresh process lerobot format
5 05_driver_real_arm.py six joints read from a physical arm, one no-motion write SO-101
6 06_teleop.py scripted leader to follower records a demonstration SO-101 / sim
7 07_mesh_two_peer.py two peers exchange presence, state and RPC Zenoh / loopback
8 08_dashboard_live.py the dashboard shows joints byte-equal to the driver's sim
9 09_agent_tools.py every tool runs, including every refusal path all
10 10_preflight_refuses.py preflight() refuses a unit-frame mismatch before motion core
11 11_mhs_device.py Robot("so101") mounts as an MHS device: manifest, act, execute_op, e-stop MHS SDK

Each script prints <NAME> OK <what it verified> or <NAME> SKIP <why>. The last line is

== v1.0 ACCEPTANCE GREEN (10 real / 1 skipped) ==

on a laptop with no arm attached; the skip is DRIVER_REAL_ARM, and STRANDS_ROBOTS_PORT=/dev/tty... with an SO-101 plugged in makes it eleven. Without the MHS SDK it reads 9 real / 2 skipped. real counts checks that ran against their backend; skipped counts checks whose backend is absent, each with its reason above. A skip is never a pass. CI runs the MuJoCo column headless (MUJOCO_GL=egl) and publishes acceptance/results.json, which is where the coverage matrix takes its marks.