This walking robot is described in detail in the papers below and can be seen walking in the movies below. The Science paper presents our main findings and results, while the ICRA proceedings paper acts as a supplement with detailed documentation of the device's design and a more thorough comparison of energy use. This project was aimed at accomplishing three goals: to show that the desirable properties of a passive machine could be transferred to an actively powered robot; to use this fact to build the most energy efficient bipedal walking robot to date; and to better understand the fundamentals of the neuromuscular coordination of human walking.

Actuating a passive walker, it turns out, can be relatively easy. The geometry and mass properties of this machine are very similar to those of its passive parent (above). In the demonstration videos below, you can see that the hips swing freely, with an angle bisection mechanism keeping the head/torso upright, and the knees swing freely until they reach full extension where a passive latch locks them in place. The only significant difference is that this robot has ankles that extend once per step to add power through push-off of the trailing leg. To maintain the energy efficiency of the passive machine, we used energy storage and actuation systems with high conversion efficiency and low peak power to produce, and rarely absorb, mechanical work.

Energy Efficiency in level-ground locomotion is measured by the dimensionless specific cost of transport, which is the amount of energy required to carry a unit weight a unit distance (Tucker 1975). The lower the cost of transport, the better the efficiency. This robot walks at 0.44 m/s, weighs 125 N (12.7 kg), and consumes 11.0 W of electricity to do so. This yields an energetic cost of transport of 0.20, an order of magnitude lower than any known walking robot. By comparison, Honda's Asimo, which is a marvel of engineering but does not utilize the passive-dynamics of its own limbs, walks at 0.44 m/s, weighs 510 N, consumes about 768 W (1 hp), and thus has a specific cost of transport of about 3.2. (See bar graph at right and table in ICRA paper for more comparisons)

Human evolution and learning have presumably selected against excessive energy use in locomotion (Alexander 2003). Humans have a specific metabolic cost of transport of about 0.2, which is very similar to our robot and much lower than other walking robots. Humans also seem to use both ankle push-off and accelerated leg swing to walk efficiently (Donelan 2002), powering schemes that are demonstrated on passive-dynamics based machines here and in Martijn Wisse's robots, respectively. Finally, passive-dynamics based robots have a gait that looks characteristically human. All of these facts strongly suggest that people utilize the natural dynamics of their limbs as they walk.

Paper: Collins, S.H., Ruina, A.L., Tedrake, R., Wisse, M. (2005) Efficient bipedal robots based on passive-dynamic Walkers, Science, 307: 1082-1085.

Proceedings: Collins, S. H., Ruina, A. (2005) A bipedal walking robot with efficient and human-like gait. In Proc. IEEE International Conference on Robotics and Automation, Barcelona, Spain, April 2005, in press.

Movies:

Related links:   Cornell site   Delft site   MIT site   Press