Passive Dynamic Walking with a four-legged kneed mechanism was first perfected in the 90’s. Since then, a number of groups have been able to recreate the mechanism and analyze it using simulations. The PDW here at Carnegie Mellon was built at an unknown time and apparently never walked. Our goal for our term project was to take this robot and make it walk using any means necessary.
We first received the robot early in the semester. The plan was to have the robot walking within a few weeks and then take it down to the motion capture lab to analyze its gait and compare it to that of human locomotion. We received supplies and material to create a ramp shortly after. We built the ramp using plywood and several 2’’ x 4’’s. We did not make the ramp have a permanent angle so that we could adjust it to try different configurations. At this point we began experimenting with throwing the robot. After many hours of throwing, catching, and falling, we noticed some trends about the robot. There were problems with the robot’s feet, stance, weighting, central axle, suction, and knees. In its initial configuration, the robot was not able to take any steps. Over the next few months we began working on all of these problems to try to get the walker to take more than two steps down the ramp.
The first area we began to work on was replacing the feet of the robot. Originally, one of the major problems with the way the robot moved was that the outside feet did not clear the ramp after the inside feet struck the ramp. We determined that this problem might be solved by adjusting the size and material of the feet. Our first change was to remove the thick rubber feet on the robot, and replace them with much thinner strips of rubber. This was a different material and we felt that trying a different surface would give us more insight into how the foot material affects the clearance. After replacing the feet, we noticed that the strips were too thin, and that the metal at the bottom of the legs was scraping the surface of the ramp. To rectify this, we created new thick rubber feet and glued the thin strips on top of these. This allowed us to keep our new surface, but not scrape against the ramp. After some trials with the new feet, we decided that there was no noticeable change to the way it walked. We decided that the original feet would be sufficient, but did however replace them with our own because the original ones were not all the same length and width. Having our own feet also allowed us to shape them in a way we felt would increase the chance of clearance. This modification also did not improve the walk, so our final choice in feet were ones very similar to the initial feet of the robot.
When attempting to make the robot stand statically on its four feet, we noticed that at any time, only three of the four feet would touch the floor. This meant that there was an uneven weight distribution and would cause the robot to veer to one direction after throwing. This was a major problem that we felt was the next priority to fix. We also noticed that left-outside leg of the robot was a different length than all the others. Upon further investigation, it became apparent that this leg had been modified and was configured slightly differently than the others. The modification was to allow the leg to be extended by about 1/8 of an inch. We were unsure why this was done, but were able to slightly improve the walker’s stance by adjusting the leg length. The major cause of the stance problem was determined to be with the feet. After being puzzled for quite some time about why the stance was wrong, we discovered that the feet were angled differently with relation to each other. We adjusted the feet so that they were angled correctly and the walker was then able to stand up on all of its legs. This is something that we constantly have to adjust to maintain the proper stance. There is a problem with the size of the holes and screws that connect the feet to the legs. This allows for room for movement, and thus gradually changes the angle of the feet every time the walker is thrown.
The next step was to determine optimal placement of external weights to allow for walking. The weights we were given to do this were not ideal. They were obscurely shaped and non-uniform chunks of metal that we placed on various locations on the walker. After doing some research on successful PDW projects, we decided that the best placement was having many weights at the top of the robot near the hip, and smaller weights on the knees of the walker, which increased the force of the swing on the outside legs. We also tried weights behind the knees and on the upper thighs. Eventually, we were given many metal balls with which we created a set of uniform weights to place on the robot. After every change we made on the robot, the weights would have to be adjusted to optimize the performance of the robot.
After some time, we noticed that there was a problem with the central axle of the walker. When one turns the axle the legs move out of proper alignment. This is probably due to some minor bending in the axle. We tried to fix this by lubricating some of the bearings, but were unable to rectify the issue. Replacing the axle would be a task that we felt we would be unable to do, and the movement of the legs was not very significant. This may be something that future people working on this robot might want to consider.
After all of the above improvements and much trial and error with throwing, we were able to get the robot to take two steps down the ramp. At this point, the legs would buckle and the robot would fall down. This was when we realized that the most important aspect of the robot’s ability to walk lied in the knees and suction system. The original configuration had suction cups attached to the upper portion of the knees with tiny holes in them. The area where they struck on the legs had two layers of tape, which allowed air to escape and the suction to be brief. This system had no method of modification. The suction cups either stuck or did nothing. The reason the knees need suction is because when one set of legs is swinging, the other needs to be rigid to maintain a working gait. With too much suction, the rigid legs never release, and with too little suction, they never stick. We decided that adjustable suction had to be implemented for the robot to ever have a chance at walking. We first experimented with different surfaces that the suction cups could strike on. This did not really improve the adjustability. After doing some research we decided to implement a suction cup valve system. We drilled holes into new cups and inserted thin plastic rods into them. We then cut small slits into the rods to allow air to flow in. We then inserted small Teflon lined screws into the rods. As you tighten the screw in the rod, more of the slit is covered, allowing for more suction. This allows for a high amount of modification to each suction cup, but it is difficult to get the same amount of suction on all four cups. After implementing this system, we again were able to have the walker move two steps, but then fall. There was not enough suction to have the legs lock when we needed them to with just the force of impact on the leg and too much suction when the legs needed to release. To solve this problem, we decided we needed to increase the moment arm to allow for higher forces on the suction cups.
In order to accomplish the above, we needed to change the knees. In the original design of the walker, the knees were three inches long. By the time we started working with the robot, these knees had become very bent. We felt that by creating new knees, we could not only solve the problem of force, but also improve on the quality of the knees. We designed new knees that would be five inches long and have the same features as the original ones. These new parts would have to be machined.
We received some stock aluminum from the Mechanical Engineering Machine Shop, and after completing the mandatory shop safety class, were able to get into the shop to fabricate the knees. We successfully machined four new knees which we feel is a major improvement to both the functionality and style of the robot.
After implementing the new knees we again set to throwing the robot with different weight configurations, ramp angle, and levels of suction over and over again. Unfortunately, we were unable to have the robot walk down the ramp.
We feel that with the right combination of weight and suction that the robot is completely capable of completing its task. We were just unable to find this setup. This might be accomplished with the aid of simulations. The amount of improvements we made to the robot was significant. With the original configuration of the robot, there was absolutely no chance of walking. We feel it now has the ability to.