MODULE 15 / MOTION & INTELLIGENCE
Run, measure, and debug a walk
Reproduce the audited walking sequence with a portable simulation-only runner. Save exact settings and compare commands with motion before changing a single parameter.
By the end, you can…
- Run the scripted fifty-second baseline
- Interpret body-frame speed and world-frame path
- Diagnose a controlled action-history regression
Inside one walking decision
Bring BDX-R into simulation
Build a complete repeatable experiment
A run is more than a policy file. It includes the robot description and meshes, dependency versions, startup state, gains, observation construction, physics step, policy period and command sequence. The course runner uses new.onnx plus bdxr.xml, a 0.005-second physics step and a policy decision every 0.02 seconds. Four physics updates occur between policy decisions.
The default script runs standing, forward walking, stopping, backward walking, turning and stopping again. It imports only the upstream configuration and simulation classes, bypasses terminal/audio startup and never imports the hardware backend. It also keeps the demonstrated previous-action correction explicit. This is an educational simulation adapter, not a robot deployment program.
A command is a request you must measure
A requested forward speed of 0.2 m/s need not produce exactly 0.2 m/s. Measure the actual body-frame velocity, sideways velocity and turning rate over a defined interval. Also inspect the time trace: an average can hide a brief failure or a long transient. The default sequence includes opposite directions, so one mean over all fifty seconds is not a useful forward-tracking score.
Body coordinates move with the robot; world coordinates belong to the room. A robot that turns while walking forward can travel strongly along world Y even when its sideways body velocity is small. To claim foot slip, you would need foot/contact measurements. A curved path by itself does not establish sliding.
Use known quirks as experiments
The earlier audit found that −0.1 m/s backward commands produced little reverse movement in several combinations, while −0.2 and −0.3 produced clear backward motion in the selected policies. That is a measured operating-range question. Reproduce it before deciding whether the policy is broken. Record the command, measured speed and transient interval used for comparison.
The old and corrected action-history orders provide another useful comparison. Change only the history mode and watch when behavior diverges. If a run fails, distinguish model loading, settling, first inference and later motion. “It fell” is a starting observation; locating the phase narrows the likely cause.
Protect the baseline from accidental improvement
Use a new output name for every run. The runner refuses to overwrite an existing CSV or JSON pair. Keep the reference source untouched, and save a diff when you edit a lesson script. If a change seems better, rerun the original case as well as the new one; an improvement in one command can make another worse.
The audit’s exact fifty-second sequence stayed upright, with maximum base tilt about 3.54 degrees. That is a reference result for one model and environment. Small numerical differences on another machine are not automatically errors. Large behavior differences deserve investigation before widening the test conditions.
WORDS YOU NOW KNOW
- Baseline
- An explicitly recorded working configuration used for comparison.
- Regression
- A previously working behavior becoming worse after a change.
- Tracking error
- The difference between requested and measured motion.
YOUR CALL
Finished with this module?
0 of 7 lab steps checked. 0 of 3 questions answered. You decide when the module is complete. Revisiting or editing it will not change your decision.
You can reopen it later. Checking every box will not automatically complete it.Go to the source
Primary references for this module. Use the version and hardware revision named in the lesson.
Pinned Deploy simulation backendPinned BDX-R training timingDownload walking labDownload graph toolDownload lab setup guide