I'm involved in research on psychologically plausible models of reinforcement learning, and as such I thought it'd be nice to try and see how well some to the models out there perform in the real world (i.e. sensory-motor learning on a mobile robot). This is already been done in some robotics labs, such Sutton's implementation of the Horde Architecture on the "Critterbot". However, these implementations involve robots custom-build by robotics experts in order to deal with the trials and tribulations of learning on a long time-scale:
"The robot has been designed to withstand the rigors of reinforcement learning experiments; it can drive into walls for hours without damage or burning out its motors, it can dock autonomously with its charging station, and it can run continuously for twelve hours without recharging."
Unfortunately I'm no expert when it comes to designing robots, and don't have access to a high quality machine shop even if I did; I'm stuck with whatever I can buy off-the-self or assemble by hand. Are these constraints common enough for amateur robotics suppliers to cater to, or should I expect to have to start from scratch?