A colleague pointed me to this very interesting robotics project undertaken by Kevin Lynch and Howie Choset. The stated objectives are:
1. To design, build, and control a planar robot to dynamically climb in a rigid vertical chute.
2. To design, build, and control a planar robot to reliably move from an initial state to any dynamically reachable goal state in a complex vertical environment of footholds at various orientations.
As someone with experience in the area of designing control strategies for complex robotic tasks, I would say that most people from outside the area tend to greatly underestimate the underlying (conceptual) difficulties – especially of point #2 which is important. As the PIs state, “the true goal of this work is not to build speciﬁc climbing robots, but rather to address scientiﬁc issues fundamental to dynamic legged locomotion in non-trivial terrains by confronting these two challenges”.
This last point is a noteworthy one. All too often, people look at a robotics project as nothing more than a demo. In the process, one risks overlooking interesting scientific questions that could spur advances in allied areas. For instance, a natural extension question – how do the results of #2 inform the design of learning algorithms for complex control strategies (novel representations, protocols, etc.)?