Debriefing
Performance
Overall, RoboPele performed well in that it could adapt to a wide
range of input speeds without much difficulty. RoboPele's cart could
hit any ball coming off the ramp; from the slowest and lowest angled
ramp to the fastest and highest angled ramp. It was disapointing that
the second sensor, even after the switch to a light sensor, would fail
at high speed. But that was only at the fastest speeds and was
approaching the maxiumum velocity that the cart could go. The two
greatest contributing factors to this was the removal of the RCX from
the cart as well as the feedback mechanism for catching slow balls
after a fast start.
Robustness
The physical robustness of RoboPele was quite admirable. Due to
initial design failures where the torque produced by two motors would
rip the cart apart after a couple level zero to level eight power
changes, subsequent designs could take survive in well in an adverse
environment. A few accidental, and later not so accidental, thigh
height drops demonstrated this execllent behaviour. The team is quite
proud of RoboPele's surviability.
Limitations
The most disappointing element of RoboPele is the temporal sensitivity
of the input sensor. Neither the light nor the touch sensors trigger
if they are not in the triggering state long enough. This of course
limited the sensitivity of the ball speed input. Additionally, there
were a limited number of sensors which could be connected with the RCX
unit. Had more of these been available as well as a greater temporal
accuracy the original design of measuring the amount of time spent
under a light sensor could have been possible. In fact with suffienct
light sensors an array of light sensors could be setup such that
RoboPele could respond to any ball coming at in angle and any
velocity.
Due to the limitations of the Lego Mindstorms programming language
there was one feedback loop the team wanted to build, but could
not. The feedback loop we wanted to build was one that would take
inputs on the sensors after a trial in such a way as to inform
RoboPele if the delay and/or colision speed of the previous trial's
ball speed was to short, too long or just right. In order to implimt
this learning feedback, RoboPele would have required variables more
advanced that Mindstorms's counter unit. Had this been available then
it would have made the process of training internal to RoboPele and a
self contained closed loop control.