Bioinspired Control Architectures for Multilegged Locomotion

Lead Research Organisation: Royal Veterinary College
Department Name: Comparative Biomedical Sciences CBS

Abstract

The primary aim of this research is to discover control principles for multi-legged locomotor systems that will allow them to move safely over soft surfaces. The secondary aim is to understand what determines the amount of energy it takes to move on these soft surfaces. The results of these aims will be synthesized to propose a control architecture that is both stable and energy efficient for multi-legged systems moving on soft surfaces. This research has broad, important applications in engineering and biology. Any robot designed to operate in the real world will innevitably encounter some form of compliant surface, upon which it must remain stable, and, if controlled properly, could reduce its energy consumption. This research would enable that robot to move safely and efficiently on soft surfaces. More agile, efficient robots have immediate and critical application to search and rescue, space exploration, and disaster relief. Furthermore, a better understanding of legged systems would enable us to build prosthetic limbs that work naturally on rough or moving terrain, and to provide better rehabilitation to those with neurological disease or injury. For biology, while we know that multilegged animals (>2 legs: the overwhelming majority of them) routinely handle compliant surfaces, we don't know how these surfaces affect their stability or energetic consumption. This means we don't understand how the environment has shaped them throughout evolutionary history. Further, their mechanisms of dexterity on compliant surfaces remain an untapped source of bioinspiration. Thus the basic scientific results of this proposal have fundamental import in robotics, medicine, and biology.To reach these aims we require three things: 1) a way to specify and systematically vary the controller in a running multi-legged system; 2) a way to measure its stability and energy consumption; and 3) a way to search through the large space of possible controllers in a reasonable amount of time. This proposal will meet these challenges with a cross-displinary approach that integrates results from robotic and animal studies. The robot provides a platform in which the controller can be directly specified and systematically changed, while the stability and energetic consumption of it's motion can be readily measured (the on-board power management circuitry gives energy consumption directly). Because there are a huge number of ways to organize the control of multiple legs, we will use measurements of how trotting dogs adjust for compliant surfaces to guide how we vary the controller in the robot. Measurements of how dogs adjust for the compliant surface will be taken back to the robot, programmed into it's controller, and the effects of making those changes quantified in a real, moving system. This overcomes the problem with animal experiments that it is not known what control architecture the animals is actually using, as well as the limitation of using computer simulations to design controllers, which is that current models of foot contact are often so unrealistic that they can't be used to predict the behaviour of a real world robot or animal.We choose dogs because they are known to be agile, efficient runners, even in the face of a dynamic environment, and past research has shown that both dogs and the legged robot RHex have similar bouncing gaits. They can be safely challenged with a soft surface, and yet are large enough to measure the stability of their body and the motion of their individual limbs. Past insect and mathematical research has suggested that by acting as a point mass with an elastic spring leg actuated at the hip (the clock-torqued spring loaded inverted pendulum, or CT-SLIP), a multilegged system may be able to handle soft surfaces with simple, feedforward control. The robot will be programmed with this controller as a starting point, and the animal (a trotting dog) is known to behave in this manner on rigid surfaces.

Planned Impact

The aim of this project is to improve our control architectures for multi-legged locomotion, by comparing perturbation experiments on a cutting edge legged robot with an extraordinarily capable natural runner, the trotting dog. These improvements in control will enable us to build more agile and economical robots. Because the UK currently lags behind the US and Japan in robotics research and commercialization, this proposal represents an important strategic investment in UK robotics. Without an improved understanding of legged locomotor control, we cannot build prosthetic devices that enable amputees to walk naturally on rough terrain, we are limited in our ability to rehabilitate those with neurological disorders, and we miss opportunities to improve animal welfare and to build agile robots that assist us in search and rescue, disaster relief, and mine clearance. The accessible yet multifaceted nature of the project make it an ideal vehicle for public engagement with science, and a valuable asset to third sector organizations such as the Science Museum. In sum, the basic scientific importance, commercial applications, and publicly accessible nature of this project give it broad impact across the academic, commercial, and public sectors. This proposal represents a timely and strategic investment in the UK's commercial sector that will bolster its economic competitiveness. The International Federation of Robotics and the United Nations Economic Commission Europe recently predicted that the robotics industry will be worth 38 billion by 2025. This proposal will engage with UK based active prosthetic device makers Chas A Blatchford, surgical robotics company Acrobot, and dextrous robot manufacturers Shadow Robot. The project is especially well matched to active prosthetic limb design, as its improvements in control architecture could make these limbs more stable and energy efficient when used to move on soft or dynamic environments. Opportunities for commercialization in prosthetics, surgical, service, and entertainment robotics will be sought with these UK robotics companies and others, through the experienced commercial applications team at the Royal Veterinary College. In the public sector, points of impact will be undertaken in medicine, animal welfare, defense, hazardous waste cleanup, and search and rescue operations. The PI's group provides biomechanics support for two of the leading orthopaedic hospitals in the UK, the Royal National Orthopaedic Hospital and the Nuffield Orthopaedic Centre. The potential for results from this project to aid in clinical gait rehabilitation will be discussed with investigators at each of these hospitals. Further, the Farm Animal Welfare Council stated in 1997 that lameness is the most important animal welfare problem in the dairy industry , with one in two dairy cattle affected by lameness each year. This proposal will shed light on which surfaces are optimal for preventing animal lameness, and will be pursued through the group's contacts built up during a recently completed, three year DEFRA project on cow lameness. The wider public will be engaged through a scientific outreach and education event, media activity, and a dedicated website, capitalising on the exciting, accessible natures of animal locomotion and robotics. An open day will be held at the Structure and Motion Laboratory, entitled Dogs and robots on springs: how many-legged animals move , including demonstrations of the robot and technologies used at the frontier of motion science. Media coverage will be exploited to publicize the project with the aid of the RVC's PR consultants. The PI has experience with and enthusiasm for engaging with the media, having recently appeared in The New York Times, Time.com, New Scientist, the Daily Telegraph, National Public Radio (US) and the TV network ESPN (US).
 
Description This grant has produced a ground-breaking mathematical model of quadrupedal locomotion. Constructed using a dynamical systems approach, fit to data from dogs changing gaits on a treadmill and then confirmed with data from dogs free running on rough terrain, the model opens the door to a host of new questions across biology, and better controllers for robots.

The controller works by placing a potential function on the space of relative leg phases, that drives the system to the idealised walk or trot gaits. As velocity is increased, the potential changes, causing the system to change gaits from walk to trot. Fits to data from dogs changing gaits on a treadmill are quite good-steps to model hysteresis, which we have planned, will produce even better fits.



Currently, the robot RHex, which this grant enabled us to purchase, does not co-regulate legs when a single leg is perturbed. We are currently applying this new potential function controller to the robot. Because it will co-regulate legs when one is perturbed, in a manner that maximises stability margin, we expect the robot to pitch, roll, and yaw less with this controller on rough terrain.



Finally, we have formulated the model so that it can have broad impact on biology. We used a dynamical systems framework to enable integration with other levels of the animal, yet chose small numbers of parameters that have direct interpretation in terms of the animals' body shape, to enable generalisation. The model allows us to answer new questions in Neuromechanics. Because it is expressed in the dynamical systems framework, it can be more readily linked to dynamical models of neural networks in the spinal cord that generate gait. Being a model of four legged locomotion, it can beused with new perturbations of running mice such as those enabled by optogenetics.



As for the search for general principles of control of animal locomotion, it enables this because the parameters in the model represent quantities such as the aspect ratio of the quadruped: is it tall and narrow, like a giraffe, or short and squat, like a sausage dog? The model makes predictions about how these animals will recover from perturbations, which we propose to test. The final three conference papers issued on this project have introduced this model; we are currently conducting new robot experiments with this controller. Once this is finished, we will have an integrative picture of animal data inspires new control architecture which makes the robot better-and having done the animal work and developed the new controller, we are currently writing this up for a high impact journal, with the robot data to be included once finished.
Exploitation Route In addition to the robotic routes to commercialisation above, the new model/controller we have developed has applications in medicine and veterinary care. We are currently fitting the model to data from walking and running mice. Because the model incorporates the full time evolution of the oscillating limbs, it is much more sensitive to changes in gait. We are seeking meetings with pharmaceutical and biomedical industry over the potential of the model in diagnosis of neurological disorders. Further, we are seeking a clinical partner to fit the model to horses with varying degrees of lameness and neurological disorder. It may have exceptional sensitivity at detecting these conditions.



It will have use in the following sectors:



Robotics:

- Research

- Toy / commercial

- Search and rescue / hazardous environments

- Defense



Biomedical

- Improved assays in rodent models of disease



Veterinary

- Improved diagnosis of lameness and neurological disorder.
Once we have tested whether the controller is more stable (and if so it is likely to be more energy efficient), then we will look for ways to get it into commercial versions of the XRL robot. We will consult with Boston Dynamics, the Kod*Lab at Penn, and manufacturers of toy quadrupedal robots, to look for routes to improving existing robots for sale (Rugged Rhex at Boston Dynamics, or Aibo from Sony).
Sectors Other

 
Description As stated above in the Research Output section, our results from moving animals, modeling, and robotics are having an impact on academia already and, we hope soon in industry. We have shown that dogs do not appear to stiffen their legs on soft surfaces: in the biomechanics community, this is changing our thinking about how many-legged animals compensate for complex substrates. The dynamical systems model of quadruped locomotion we have developed will be critical in understanding how four legged animals of different size control their legs, how spinal interneurons produce coordinated leg motion, and is likely to produce more stable robots that can change gaits (walking to trotting) simply and elegantly in the same manner as animals. These have the potential to be huge impacts in biomechanics, neuroscience, and robotics. The fact that we reached 8000 public visitors at the science museum with a dynamic running legged robot shows is important impact for STEM education in the UK. Our proudest moment came from a girl who ran up the robot and said "Wow! I've just come from the natural history museum, and I have to say: this is COOLER than the dinosaurs!" The PI has met briefly with the CEO of Shadow Robot in London, and will be pursuing the contact to look for future fruitfly lines of industrial collaboration. . Our results from moving animals, modeling, and robotics are having an impact on academia already and, we hope soon in industry. Our continuing research has the potential to have a huge impact in biomechanics, neuroscience, and robotics areas. Beneficiaries: Academia. Contribution Method: We have shown that dogs do not appear to stiffen their legs on soft surfaces. This is changing thinking about how many-legged animals compensate for complex substrates. The dynamical systems model of quadruped locomotion will be critical in understanding how four legged animals of different size control their legs and how spinal interneurons produce coordinated leg motion, resulting in more stable robots that can change gaits simply and elegantly in the same manner as animals.
Sector Other
Impact Types Cultural

 
Description Inspiring young people to STEM education
Geographic Reach National 
Policy Influence Type Influenced training of practitioners or researchers
URL http://www.vmine.net/scienceinparliament/sip.asp
 
Description Army Research Office - Single Investigator Award
Amount $390,000 (USD)
Organisation US Army Research Lab 
Department Army Research Office
Sector Public
Country United States
Start 05/2014 
End 04/2017
 
Description Royal Society Research Grant
Amount £13,900 (GBP)
Funding ID RG100633 
Organisation The Royal Society 
Sector Academic/University
Country United Kingdom
Start 05/2011 
End 06/2012
 
Description Kod*Lab Collaboration 
Organisation University of Pennsylvania
Country United States 
Sector Academic/University 
PI Contribution The main collaboration of this grant is with Prof. Daniel Koditschek, chair of engineering at Penn, in the USA. This collaboration has been extraordinarily fruitfull. The broad impact of the research output above was made possible by the integrative nature of this collaboration. Postdocs G. Clark Haynes and Shai Revzen, undergraduate student Matthew Hale, and PhD student Aaron Johnson of the Kod*Lab at Penn have made outstanding contributions to this grant. Through a series of four visits (2 by RVC to Penn, 2 by Penn to RVC), the robot was delivered to the RVC and used to conduct several exciting new experiments, and the RVC team was able to travel to Penn to get trained in the robot?s programming, but more importantly, to inspire new directions in the bimoechanics and the control of locomotion. The synthesis of the RVC teams? expertise in gathering data from moving animals, with the applied mathematics and robotics expertise of Penn, has been extremely productive. The Penn team suggested experiments involving gait transitions in dogs, and using rough terrain locomotion as a method of injecting noise into quadrupedal locomotion. The RVC PDRA on the project has, with guidance from Prof. Koditschek and the Kod*Lab team, developed the potentially very general dynamical systems controller for quadrupeds. This control is leaps and bounds ahead of the foremost past theoretical effort, of Schoner et al., J. Theoretical Biology 1990.
Start Year 2010
 
Description EPSRC Medical, Mechanical and Materials Advisory Committee 
Form Of Engagement Activity A formal working group, expert panel or dialogue
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Professional Practitioners
Results and Impact The PI was invited to sit on the EPSRC Medical, Mechanical, and Materials Advisory Panel on Feb 16th, 2011. Involvement in advisory committee with national remit

Aided in grant award process.
Year(s) Of Engagement Activity 2011
 
Description Harvard University seminar 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact "Insects on rubber and dogs on springs: sensing and perturbing animals to understand the mechanics of legged locomotion." CFS Seminar, Dept. of Organismal and Evolutionary Biology, Harvard University, Massachusetts, March 11th, 2011.

INcreased awareness of research program internationaly.
Year(s) Of Engagement Activity 2011
 
Description Northeastern University, Boston, USA 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact "Insects on rubber and dogs on springs: sensing and perturbing animals to understand the mechanics of legged locomotion." - Boston Action Club, Dept. of Kinesiology, Northeastern University, Boston, MA, March 10th, 2011.

Increased awareness of research program internationally.
Year(s) Of Engagement Activity 2011
 
Description Robotville Festival 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Public/other audiences
Results and Impact The outreach work with the robot has been extraordinary. We were invited to take part in the Robotville Festival at the Science Museum in London. We demonstrated the running, rubble conquering Rhex Legged robot to nearly 8000 visitors over four full days, and PI Andrew Spence gave two public outreach talks during these visits. This included a visit on the Sunday by Leader of the Labour Party, Ed Milliband. The event was covered live on national television, and garnered very wide publicity.

Over 20 unique robots set up camp at Robotville at Science Museum. A four day festival packed full of special events and exhibits, Robotville explored the cultural significance of robots and gave attendees the chance to see first hand the most innovative and exciting developments in European robot design

We had 8000 visitors including Ed Milliband who were made aware of the interest and importance of bioinspired robotics.
Year(s) Of Engagement Activity 2011
 
Description Starting an independent research career 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Professional Practitioners
Results and Impact Andrew Spence was invited to give a seminar on "Starting an independent research career" at UCL. His talk was part of the UCL Neuroscience Domain - Early Career Forum and was part of a series of talks for postdocs and junior PIs on starting an a career in academia.

Audience seemed interested to hear about getting a career started.
Year(s) Of Engagement Activity 2011
 
Description Temple University, Philadelphia 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact "Insects on rubber and dogs on springs: sensing and perturbing animals to understand the mechanics of legged locomotion."

Departmental seminar, Dept. of Biology, Temple University, Philadelphia, January 13th, 2011.

Increased awareness of resarch program intenrationaly.
Year(s) Of Engagement Activity 2011