04 / Experiments
Compare tracking, terrain response, and caster motion in simulation and on the real platform.
Experimental setup
ROS 2 Jazzy and Gazebo Harmonic run the ball and wheelchair caster configurations over the same route. Record ground-truth pose, tracking error, joint motion, and commands.
Swap the support casters, preserve the payload and controller settings, and repeat the marked route. Synchronize pose, wheel and caster encoders, IMU, video, and force measurements.
The current simulation shares chassis geometry and commands; caster mass, contact, and bearing parameters are model assumptions. Hardware calibration and repeated trials will support the final comparison.
01 / Trajectory tracking
An archived pair of simulation runs on the ADTG-d25 terrain. Select a configuration or replay its trajectory history.
Values belong to this single simulation pair. They are not an aggregate or a real-world result.
RMS = √mean((x − xref)² + (y − yref)²), evaluated against the timed reference. The plot shows recorded samples; the displayed RMS values come from the full run summaries. Ball CSV · Wheelchair CSV
| Condition | Ball caster RMS | Wheelchair RMS | Trials |
|---|---|---|---|
| Simulation · final evaluation | — | — | — |
| Real world · flat ground | — | — | — |
| Real world · uneven ground | — | — | — |
02 / Off-road terrain
Uneven support changes contact, chassis attitude, and tracking. Paired routes let us measure those effects without changing the reference.
Both runs complete this route. The ball configuration has slightly lower position RMS in this pair; broader conclusions need repeated tests across terrain and contact settings. These rigid surfaces do not model soil deformation.
Add the side-by-side video, measured terrain profile, and a one-sentence finding here.
03 / Disturbance & resistance
A wheelchair caster must swivel its fork into a new rolling direction. The split ball mechanism can accommodate changes through its shell motion; the experiment tests whether that reduces reorientation resistance and path deviation.
Apply the same timed lateral push at the same chassis point, then measure peak tracking error and recovery time. In a separate pull test, use a force gauge to measure the force needed to start and sustain motion.
Lower pull force and stronger disturbance rejection are separate measurements. Record both before drawing a performance conclusion.
| Measurement | Ball caster | Wheelchair caster |
|---|---|---|
| Applied lateral force / duration | — N / — s | — N / — s |
| Peak / steady pull force | — N / — N | — N / — N |
| Peak path error / recovery time | — m / — s | — m / — s |
04 / Caster motion analysis
Caster Vision provides an additional estimate of caster orientation: fork markers and wheel texture for the wheelchair caster, and tracked surface features for the ball hemispheres. Synchronize these estimates with encoder logs to compare the motion over a trajectory.
Replace with the synchronized result image, including valid tracking intervals and the angular error.