Teaching robots to run like birds: Comparative biomechanics inspire novel robot control [with video]

Robotic technologies have experienced a renaissance in recent years— yet the best humanoid robots remain far less capable of dynamic balance compared to humans and animals. We have discovered a simple control principle, inspired from running guinea fowl, which may help bipedal robots achieve the robust and agile athletic performance of these remarkable animals.

HFSP Young Investigator Grant holders Monica Daley and Jonathan Hurst and colleagues
authored on Mon, 03 February 2014

The helmeted guinea fowl, a chicken sized bird from Africa, may seem like an unimpressive critter.  Yet, these ground birds provide an ideal animal model for understanding principles of bipedal locomotor control. Guinea fowl move with speed, economy and agility through complex natural terrain environments, and can travel up to 20-30 miles a day while foraging.  Inspired by the athletic prowess of guinea fowl, we formed a team of physiologists and robotics engineers to rigorously study the control rules they use to achieve dynamically stable walking and running gaits. This work has been funded by a HFSP Young Investigator Grant.

Figure:  Guinea fowl with the robot, ATRIAS. Photo credits to Dr Alexander Spröwitz.

Walking and running are such common daily activities— we might superficially think they are simple and easily understood. Yet, the ability of current robotic and prosthetic devices to navigate complex terrain remains frustratingly limited. Further, anyone who has watched a toddler practice walking and running can appreciate that these tasks are very challenging, requiring the precise coordination of neural and motor systems to achieve safe foot placement and balance. Consider then, that many cursorial ground birds, like quail, guinea fowl, emus and ostriches, master walking and running immediately after birth. This suggests that much of the complexity of dynamic gaits can be ‘hard-wired’ or ‘pre-programmed’ and doesn’t necessarily require the extensive learning process observed in humans. 

We set out to understand the control rules used by guinea fowl to achieve balance during dynamic running gaits, through combinations of experimental terrain perturbations and computer simulations. In computer simulations, we have used an approach that separately considers 1) the mechanical system, 2) the control applied to the mechanical system, and 3) the performance priorities for which the control is optimized, such as economy, steady gait or injury avoidance.  By optimizing the model’s control for different possible priorities and comparing predictions to guinea fowl behaviour, we can infer some of the ‘rules’ governing their leg control during locomotion.

Robotics engineers often consider ‘theoretical’ notions of stability that do not apply well to human and animal locomotion. One reason these theories fail is that they do not explicitly consider the mechanical demands and specific failure limits that influence animal behavior.  Yet, the physical interaction with the environment is fundamental to how animals move.  These physical interactions determine the energy cost and injury risks of movement, and directly influence maximum speed and maneuvering performance.  Animals must be able to move quickly through complex terrain to avoid predation, without breaking a leg if they stumble.  Similarly, humans must be able to navigate complex terrain with curbs, stairs and unexpected potholes without injury.  Current theoretical approaches to stability in robots do not adequately address these issues.

Our recent paper presents a new bio-inspired model for leg control of running that explicitly considers these mechanical demands.  In this paper, we make predictions for optimal swing leg control based on bird-inspired performance priorities— energy economy and injury avoidance. We compare simulations based on bird-inspired priorities to those based on more traditional engineering notions of stability. We find, serendipitously, that energy economy and injury avoidance are closely aligned priorities, allowing them to be achieved simultaneously using a simple swing leg control policy that can be easily ‘pre-programmed’ into a robot.

We further show that this bio-inspired control conflicts with a typical engineering approach. Robotics engineers tend to directly target stability, controlling the leg to minimize deviations from steady gait. However, we find that this strategy has some undesirable mechanical consequences, increasing leg loading and leg mechanical work in uneven terrain, which incurs increased energy cost and risk of injury. In contrast, we find that our bird-inspired control can simultaneously achieve economy and safety, while also maintaining sufficient stability to avoid falls. Our findings suggest a novel control policy for robots that will improve performance in uneven terrain.  In ongoing work we are continuing to analyze bird running to understand stance-phase control, and implementing our bio-inspired control on the robot ATRIAS (see http://mime.oregonstate.edu/research/drl/ ). 

Video credit to Dr Yvonne Blum.

Reference

Bio-inspired swing leg control for spring-mass robots running on ground with unexpected height disturbance. Vejdani, Blum, Daley and Hurst (2013). Bioinspiration & Biomimetics. 8 046006 doi:10.1088/1748-3182/8/4/046006.

Link to article

Pubmed link