Experiments finished, now on to the dissertation
I've wrapped up the experimental work with the instrumented bicycle and am now fast tracking the completion of my dissertation and degree.
During September, I set up the final experiments for our rider control experiments (I've unfortunately had to nix the flexible rider stuff due to time constraints). We studied the data from the first round of trial experiments and tested out some ideas on system identification. Things looked pretty good, so I fixed a few things on the bike and got three subjects to perform a series of perturbation experiments on the treadmill and the gymnasium floor. My last run number was 706 and I think that somewhere around 400 of the runs are good for analysis. This is a list of some of the improvements:
- We locked the steer torque slip clutch. This thing was such a waste of money and hassle. After a huge headache trying to get the correct clutch from the distributor, the device ended up having slop (backlash) issues. I add some shim material to take it up, but it still didn't quite work. So for the latest experiments we took the chance that the sensor could fail and locked the device so that it had no slop.
- We improved the lateral force impulse disturbance by creating a rod that we could both push and pull on. I also wrote a simple little program that provided instructions of when and which direction to apply the lateral force, both being randomized. Also readjusted the gains in the amp box so that we could get a full +/-100 lbs of force.
- The speed-o-meter was remounted so that it was tangential to the wheel.
We then went for two session at the treadmill and two sessions at the gym (it would have only been one session at the gym but I attempted to use a quickly written and test reworking of BicycleDAQ, which totally screwed up the time sycronization, and we were unknowingly over saturating the force sensor during many of the runs). After the 4 sessions with help from many folks (Mont, Luke, Charlie, Zac, Jan, Alvin, Tai, Alec, Andrew), we got the dataset which includes lateral disturbance runs for three riders, at three speeds on both the treadmill and gym floor for both balancing and line tracking. And we have the control experiments with no disturbances. We also threw in a couple of blind runs for fun too.
Peter de Lange is finishing up his master's thesis at TU Delft and was around during our experiments earlier in the year. He's attempting to use system id to fit a control model to the data we collected in a similar fashion as us. He's been making good progress (more than us!) and is currently working with the new data. We've also been trying out various system id methods with some pretty good results. We've been able to identify the parameters of our control model. I've got one more analysis that I want to perform on the data to include in my dissertation:
- Identify the 5 gains and neuromuscular frequency for each run (or each perturbation) using a parametric system id method.
- Use a linear mixed effects stats model to understand how the five parameters vary with speed and environment.
And speaking of my dissertation, I've finally started writing the real thing. Mont, Ron and Arend have all agreed to be on my committee and I can't think of a better group. I'm mostly sourcing text from in the many conferences papers, journal papers, and web writing that I already have to fill up the chapters, but there is a fair amount of new work to put in too. I'm also publishing the dissertation as a website as I write it and would love comments, reviews, and suggestions as I go along. So if you are bored and want to read a dissertation draft, feel free to check it out at: http://moorepants.github.com/dissertation/.