Post date: Jul 24, 2014 4:28:34 PM

After becoming exceedingly tired of moving individual sliders for each joint for each frame of a walking cycle I started thinking about better ways to control a walking robot. A couple of years ago I had tried using a Kinect to do real-time skeletal extraction to control FOBO (the fourth prototype). While I could get this approach to recognize large gestures, it was never close to accurate or fast enough to do real-time joint level actuation for a complicated dynamical system. Around that same time I also looked into using potentiometers and a harness on my legs to extract the joint positions. This approach showed promise because the joint positions could be read very quickly, but the harness was flimsy, didn't give consistent results, and required a lot of calibration every time it was used. Recently when I started thinking about this problem again I did some analysis on why my original harness had failed and realized that the harness needed to be rigid and that the potentiometers had to be isolated from any joint torque to keep them from failing. When I made the original harness I didn't really have the option to 3D print a rigid harness because my printer at the time (Makerbot Thingomatic) wasn't exactly capable of making the size and volume of parts needed for such a large build. My current printer, the Replicator2, is a real work horse and has been printing like a champ for over 2 years now, so I was able to move forward with fully engineered rigid version of the exoskeleton. The picture shows the exoskeleton hanging on its docking station and there is a quarter near the top for scale. The exoskeleton is made of 3D printed parts (white) and PVC sprinkler pipe (grey), and an Arduino Mega for reading the sensors and external communication. The legs and waist have latching straps to hold them in place while the exoskeleton is being worn. It currently has 10 joints each with a torque isolated potentiometer to measure the rotation. I'm planning on adding at least another joint in each foot for the toes and maybe another in the legs for yaw (although this one will be a bit tricky). I haven't tested reading all of the joints together yet, but some initial experiments look promising.