MSc Human and Biological Robotics (BHM1)
Human Neuromechanical Control and Learning (PG)
This course will provide a comprehensive and rigorous treatment of the control of human movement from the perspective of both adaptation of the neural control system and adaptation of properties of the mechanical plant, incorporating approaches from physiology, engineering and computational neuroscience. Failure to consider both neural and mechanical contributions to observed behaviour can lead to erroneous conclusions. For instance, one finds examples in the literature where effects of dynamics and muscle mechanics are wrongly attributed to neural control. This is why we and others have developed a synthesis of musculoskeletal biomechanics and neural control over the past 30 years, which we call Human Robotics. Why this name? We use the framework of robotics to understand the control problems that are being solved by the human motor system, and the insights gained by this approach lead to more versatile robots with human-like capabilities.
This course will present a comprehensive approach to understanding human motor control beginning with muscles and progressing in a logical manner to behaviour, where modelling is based on evidence from published experiments, employing a method reminiscent of physics. We will first study muscle mechanics and the properties of their sensory receptors; then pass to the mechanics of one-joint systems such as wrist flexion/extension while integrating feedback control from simple reflexes; then extend these principles to the multi-joint multi-muscle system, considering the coupled and nonlinear dynamics of the redundant musculoskeletal system; then treat motion control and planning, including motor learning during adaptation to novel dynamic environments, integrating novel computational theories based on robotics and stochastic control. The course will include several applications of the modelling, from flexible control in robotics to neurorehabilitation.
Knowledge and Understanding
Neuromechanics of human motor control and learning.
Modeling of human motor control and biological dynamic systems in general.
Modeling of biological systems dynamics and control.
Introduction : A synthesis of musculoskeletal biomechanics and neural control; Modern experimental tools; Neural organization; Control concepts.
Muscle mechanics and control: Sarcomere as building block of the actuator; Force development in a muscle; Muscle spring-like properties; Sensory receptors.
Single joint neuromechanics: The stretch reflex; Single-joint posture and elasticity control; Motion control; Linear robot control and single-joint model; Modulation of viscoelasticity; Reflex muscle activation.
Multi-joint multi-muscle kinematics: Task, joint and muscle spaces; Kinematic description and transformations; Mechanical impedance geometry under static conditions; Redundancy in the human arm and in robots; Optimal configurations.
Multi-joint dynamics and control: Action dynamics; Linear and nonlinear control; Force and impedance during arm or leg movement; Dynamic redundancy and applications for animats.
Motor learning and memory: Adaptation to a force field: emergent model properties; Modeling learning of stable dynamics in humans and robots; Generalization in motor learning; Motor memory and consolidation.
Motor learning under unstable conditions : Motor noise and variability; Learning optimal impedance for unstable and unpredictable dynamics; Computational model of human motor adaptation.
Motion planning and online control: Optimal movements; Minimization of error and effort; Stochastic optimal control modeling; Synergies and motor primitives; Visuo-motor coordination; Planning in the presence of multiple solutions.
Integration and Control of Sensory Feedback: Proprioceptive and visual feedback; Adaptive control of feedback pathways; Sensor fusion and Bayesian formalism; Forward model; Purposeful vision and active sensing.
Neurorehabilitation: Stroke symptoms and importance; functional and sensor-based assessment of the sensory-motor function; rehabilitation robots; adaptive training modes; rehabilitation neuroscience.
The course is for students with background in computer science, mechanical, electrical and bioengineering, as well as psychology, kinesiology, rehabilitation therapy, neurology, physics, who are interested in understanding the algorithms of human motor control. The material is self-contained so that it can be understood by these distinct communities with their different perspectives and knowledge. Linear differential equations, linear algebra.
Students will be taught over one term using a combination of lectures and study groups. Lecture sessions will be made available on Panopto for review and supplemented with technologies to promote active engagement during the lecture such as 'learning catalytics'. Study groups will be based on taught content from lectures to reinforce these topics and allow students to test their understanding.
Lectures: 15 hours
Labs: 6 hours
Study groups: 12 hours
Workshop: 2 hours
● Written exam: ; % weighting
Rubrics: The exam paper will contain 3 questions, which have to be answered in 2 hours. These understanding questions are conceived so that they can be answered by people from different background, so either with physiology or engineering background.
Outline answers to past papers will be available
● Written report: Tutorial 1; 4% weighting
● Written report: Tutorial 2; 4% weighting
● Written report: Tutorial 3; 4% weighting
● Written report: Tutorial 4; 4% weighting
● Written report: Tutorial 5; 4% weighting
● Written report: Tutorial 5; 4% weighting
Feedback : The corrected five tutorials sheets will give feedback on the learning and be given back to the students 2 weeks after submission.