robotics test bench / experiments
MuJoCo experiments in feedback control, coupled dynamics, contact, and leg reachability.
latest notebook / 06
the two-joint leg misses the chosen support point by about 6.7 cm. a constrained solve with an added joint reaches the target to the printed precision. support under load remains untested.
open the notebook ↗read experiment 06 →
Can the current C-1N leg reach an outward-and-down support point?
the two-joint leg misses the target by about 6.7 cm. in the constrained solve, an added joint reaches it to the printed precision. that does not yet establish support under load.
What makes a tripod pose statically supported?
the first sweep used an incorrect ballast configuration and was stopped. a corrected fixed-foot sweep located the boundary near zero support margin and rear-contact unloading.
Why can the same joint command move a foot differently?
pose-dependent joint-to-foot velocity mappings and the corresponding force-to-torque calculation.
What does dynamics knowledge buy, and what happens when it is wrong?
feedforward reduced tracking error when the model was accurate. changing only the controller’s mass estimate brought some error back.
What happens at one joint when only the other joint is driven?
the elbow moved even with zero elbow torque. driving one joint changes the motion of the whole arm.
What does closing the loop change in a one-joint pendulum?
P-only control overshot or oscillated. adding velocity feedback helped the pendulum settle.
C-1N STRIDE checkpoint evidence ↗
ongoing lab work now lives in the C-1N repository. the six published studies below remain historical records.