humanoid leg joint actuator load

[_ADAM_]

I've been thinking a lot about humanoids lately, and I'm wondering how load is distributed among the leg joint actuators while the robot is both standing and walking.

For the sake of discussion, assume this 6-DOF leg design:
http://www.trossenrobotics.com/p/cus...Robot-Leg.aspx

If anyone has done this analysis or knows of any research, I would appreciate the information.

-Adam

[tician]

Depends on the gaits and assorted motions you make the legs perform. Longer segments between joints mean larger torques when trying to stand from a sitting/squatting pose, but can make walking more efficient since it may permit a larger and slower leg swing during each step. A bit of reading about statics and free-body diagrams would be most helpful for determining feasibility of the combination of segment lengths, servo torques (continuous torque of hobby/robot servos is often ~1/5 of stall torque), and bot poses. Statics is one of the things thrown at first year mechanical/structural engineering and physics students, so not too difficult to do a basic analysis of a bot's pose and some CAD/CAM programs might be able to help a bit (definitely help with FEM/FEA analysis of parts under loading).

When using a human-like gait, the ankle+knee+hip really only significantly bend when the foot is off the ground while singly supported by the other leg which is essentially straight with the knee locked and tilt controlled by ankle+hip. This means the load on most of the servos in both legs is quite small: support leg is in pose that requires little torque in most of the servos to maintain and lifted/swinging leg is only moving its own mass.

If using something like the DARwIn-OP gait with the knees always bent, the torque requirements can go up significantly. Just standing with the knees bent is likely to cause overheating after a short while and prolonged walking can require active cooling of the knees. All Dynamixel servos, except the Dynamixel Pro line, use brushed DC motors with precious metal brushes, so holding a high-load position, or otherwise stalling the motor, can result in high current with lots of heating and possible damage to the brushes and/or armature before the over-current/over-temperature safeties kick in. Some here have added small fans on 3D printed frames attached to the more heavily loaded servos and some further modifying their AX-12 to expose the motor to direct airflow without compromising the structural integrity of the plastic case. We had a cheap 8~10" fan in the lab to cool the legs of the DARwIn-OPs between demos and a CO2 duster for quicker cooling of the partly exposed motors in the MX-28T. Lost several MX-28T to over-current failures because the theatre students abused the bot and the servos were too slow to cut power to the motor resulting in damage to the brushes and/or armature. Damaged servos would be 'sticky' when rotated by hand, pull quite a bit more current when idle/unloaded (>100mA), and possibly damage other servos on the buss before their own H-bridge detonates during its next stall.

[_ADAM_]

Thanks for the info. You've given me a lot to think about.

I'm not a fan of the bent knee gaits. Doesn't look realistic. And it doesn't sound like it's good at all for knee servos. Sounds kind of scary actually. You've given me visions of flaming robot knees...

But I guess there must be some practical reasons why that gait is so common on humanoid bots.

How do the AX-12s survive in the knee join of the HR-OS1?

[tician]

You've given me visions of flaming robot knees...

Just little puffs of magic smoke, and maybe a rapidly heated servo case. AX-12 are much more resistant to damage of all sorts, although they still wind up with damaged gearsets during major collisions. In a lab of a couple hundred AX-12 (both '+' and 'A' versions) (10+ carbots with 4 each, 10+ Gerwalks with 7 each, several walkers with 6~14 each, several Type-A humanoids with 20 each, several PhantomX Reactor pinchers with 7~8 each) abused by undergrads and occasional 4~12 grade students (DUKE TIP courses), we only ever had one AX-12+ with a bad motor and/or H-bridge. The three DARwIn-OP in the lab wound up with quite a few damaged MX-28T (gearsets, motors, H-bridges, and dxl comm buffers).

Not really sure how the slightly modified DARwIn-OP gait works long-term with the HROS-1 design. Ripley received some significant damage very early on (a few bent frames, three damaged gearsets (distribution of damage meant only two replacement gearsets needed), and a broken servo case; first time I've ever heard of case damage occurring in any dynamixels), so she never spent much time as a true HROS-1 except while the folks at Trossen were testing assembly. The resident RoboEmpress would probably be much more insightful about the reliability of the two HROS models.

[r3n33]

Whoops.. I apparently wasn't lurking hard enough

So that 6 dof leg design is currently used on the HR-OS5. Those MX-106 servos have no trouble holding a bent knee pose all day on this size robot if they must. Timmy was standing continuously and actively walking on and off for all three days of RoboGames.

The HR-OS1 leg design is very similar but the AX-12 servos are operating closer to their maximum load, especially so if the robot is wearing the full armor set. At one point I tried AX-18 servos but those required active cooling to hold up to the OS1 weight and the Darwin gait for extended use ( > 15-20 min ).

As far as doing any analysis on the load distribution I have not done that specifically. I could take some measurements based on the current load register of the MX servos in different poses if you'd like. I'd even consider the code changes to request and log this data during the gait cycle with the right motivation. Kinda tempting. That information may help during the walk gait tuning process. Although at some point I can see myself moving away from the Darwin/HROS framework dependency for walking and action playback in ROS.

[_ADAM_]

Thanks for the reply.

A more specific question would be - Do you think the MX-64 would be near max load in an HR-OS5 leg? Overall robot size similar to the HR-OS5 but no armor. It sounds like the knee is the only joint that may potentially be an issue.

I think that would be a cool bit of code to have. I'm going develop something similar, but I'm prototyping with a quad first. I'll post about it when I get a bit further along.

[jwatte]

If you know that you'll be using a bent knee design, you can put a torsion spring in the knee, that provides help for the servo.
I believe there's a reason they put MX-106 in the knees of HR-OS5 other than just "let's spend more money!!!"

[_ADAM_]

I'm deliberately not using a bent knee design. I'd rather not have to put 106s in the knees, but I would if I felt like I have to. I'd rather go with all 64s. I think it will be fine. I know you guys will say to do some static analysis, but "ain't nobody got time for that."

I'd be satisfied enough to go forward if it will "probably" be ok.

[r3n33]

I think you guys are right, the MX-64s wouldn't work in the knees of an OS5 size/weight robot.

So because it seemed like fun I gathered some data from a couple of poses and while the Darwin walk gait was active. Specifically the Consuming Current ( register 68 ) of servos 9 through 18 ( Hip Roll to Ankle Roll ).

Walk Ready ( Knees Bent )
Servo: 9, mA: 0.00, Servo: 10, mA: -4.50, Servo: 11, mA: 0.00, Servo: 12, mA: 0.00, Servo: 13, mA: -341.19, Servo: 14, mA: 403.62, Servo: 15, mA: -4.50, Servo: 16, mA: 0.00, Servo: 17, mA: -0.71, Servo: 18, mA: 4.55

Standing ( Legs Straight )
Servo: 9, mA: 10.77, Servo: 10, mA: -17.91, Servo: 11, mA: -4.50, Servo: 12, mA: 0.00, Servo: 13, mA: -4.50, Servo: 14, mA: 33.03, Servo: 15, mA: 0.00, Servo: 16, mA: -28.95, Servo: 17, mA: 0.00, Servo: 18, mA: -0.00

Walk Gait Active CSV

Walk Gait Active with Body Lift CSV

[_ADAM_]

They're just numbers to me. What does it mean?