Robosapien V2: A Closer Look
Over the past six months, I had been tinkering with the Robosapien V2 to explore its inner workings. I wanted to learn how its master designer, Mark Tilden, was able to cast a life-like behavior over this hunk of black and white plastic figurine that vaguely resembles a Stormtrooper.
My initial experience was not pleasant. Two weeks after I won my prized robot from eBay, it began to work intermittently– shutting down unexpectedly when it bent down or sat. A quick Web search revealed that my Robosapien V2 suffered the fate of many of his kind because of a serious manufacturing flaw. His internal wires are insulated with cheap plastic that crumbled over time, causing the exposed conductors to short when pressed together by certain robot movements. When this happens the robot would freeze, shut down completely or in some “rare” cases, terminate itself in a toasty finale.
It takes an inspired roboticist to overlook this flaw, and thankfully I possessed that drive. I fixed the problem by painstakingly replacing each broken wire and continued with the unraveling. In a span of several weeks, I studied the RS V2′s mechanical design, probed its circuit boards, inserted an Arduino microcontroller inside its body to control him by computer, tried out the IR (infrared) control codes and carefully observed his movements. From this labor of love, I gained some insight on its ingenious design.
It seems that the secret sauce is in the way Mark Tilden creatively combined cute looks, whimsical moves, believable eye gaze using blue LEDs and a playful voice to create the illusion of a playful persona. From this concoction, we are tricked into believing that the mound of moving plastic parts is alive– in the same way that Wall-E and Eva become living characters in the Disney-Pixar animated movie “Wall-E“. On this count, the RS V2 gets high marks.
Wall-E and Eva meet for the first time.
But how well does the Robosapien V2 fare as a robot platform? The answer of course depends on what specific functions you want your robot to do. In my case, I wanted to use the RS V2 to roam autonomously inside a house (flat surface) and pick up small objects. On this basis, I found the Robosapien V2 lacking on several fronts:
- No command for reading sensors. Surprisingly you cannot query the RS V2 to determine how the head, torso and limbs are posed. The sensors are there but there is no command to read their values. Without this, it is difficult to make the robot do precise movements because you need sensor feedback to adjust for motor errors.
- The RS V2 ignores new commands when busy. Its brain chip can only process one command at a time (non-multitasking) but more than that, it discards any new commands it receives while busy processing the current command. This means you have to time your sequence of commands just right so none of them get skipped. This limitation also means you cannot run two motors simultanously, such as bending the body forward while tilting the head up.
- Not all movements can be made using the command set. For example, if you want the robot to do a partial leg split (to reach an object on the floor), there is no command to do so.
- Excessive foot slippage when walking. This makes it hard to estimate the robot’s position, and hence any type of path planning becomes really difficult.
- Limited dexterity. I did not realize how limited the Robosapien’s arm movements are until I played with them. Basically you can’t pull arms closer together to hold a bigger object (eg, a box) on its sides. Also because of its mechanical design, only hand can grasp at any time. When one hand closes, the other one opens and vice versa.
Because of these limitations, I decided to halt further development on my Robosapien V2. I considered the idea of bypassing its brain chip and using a different processor, one that I can program, to directly control the motors and read the sensors. This of course involves nothing less than performing a complete lobotomy. In the end, I scrubbed the idea because I didn’t think that I could overcome the last two problems. Perhaps someone with greater skills can solve them. But in my view, the path to greater progress is to savor the knowledge learned and move on to a more suitable robot platform.
Robosapien Explored
While working on this project, I found a few technical papers describing some interesting work involving the Robosapien. Unfortunatelly, all of them used the orignal and less-capable model, not the Robosapien V2.
Leland (2006) used a network of low-power ZigBee radios with a particle filter algorithm to compute the robot’s position. Cisternino (2005) developed the Robotics4.Net toolkit for programming various entry-level consumer robots including the Robosapien. For all you soccer fans, Behnke (2005) and team at the University of Freiburg competed at the RoboCup 2004 competition using a modified Robosapien. What’s cool about their work is that they mounted a small 1.3M-pixel camera on the robot and developed a color-processing algorithm to enable the Robosapien to navigate the field, track the ball and perform other behaviors.
NimbRo RS standing proudly at RoboCup 2004
Links
Cisternino, A., D. Colombo, G. Ennas and D. Picciaia (2005). Robotics4.NET: software body for controlling robots, in IEEE Software Proceedings, Vol 152 Issue 5. [ pdf - this is the position paper ]
Leland, E., Jenkins, O. C., and Bradford, K. (2006). Robosapien Localization and Control. Circuit Cellar Magazine, 188. [ pdf ]
Sven Behnke, Tobias Langner, J¨urgen M¨uller, Holger Neub and Michael Schreiber. (2005). NimbRo RS: A Low-Cost Autonomous Humanoid Robot for Multi-Agent Research. [ pdf ]
Credit to 
