Motor planning explains human behaviour in tasks with multiple solutions

https://doi.org/10.1016/j.robot.2012.09.024Get rights and content

Abstract

How does the human central nervous system generate appropriate movements to accomplish complex motor tasks, while adapting to the changing environmental conditions? The traditional motor control research has focused on simplified tasks in which healthy human subjects were found to utilise a single solution. In this chapter we present recent works that have examined motor control and learning in richer and more natural environments. Unlike the simplified tasks, humans use multiple solutions to solve these tasks, often switching between these solutions within trials. We exhibit the requirement of an explicit planning stage in sensorimotor control models to explain these behaviours. Behavioural experiments investigating motor planning in humans are necessary to understand all the aspects of the non-conventional control and optimisation used by humans, which will also help robots to work in complex environments such as those humans live in.

Highlights

► Recent works have examined motor control and learning in natural tasks. ► Human behaviours are found to be very unorthodox in these tasks. ► We exhibit how these behaviours elucidate the presence of motor planning in humans. ► Understanding human motor planning is essential for human like adaptability in robots.

Section snippets

Optimisation framework modelling of human motor control

Sitting at a bar, you can pick up your drink in many different ways, e.g. using a curved or straight hand movement, with a fast or slow movement, or using various combinations of muscle activations. How does your central nervous system (CNS) decide which muscles to use, and hence which hand trajectory to take in order to achieve this task? This is the degrees-of-freedom problem, which was sharply formulated by Bernstein [1]. Human motor control modelling has investigated this question for

Are current models biased by the experimental paradigms?

While the current popular models can explain most of the observations in previous experiments, probably due to logistics and for ease of analysis, almost all the previous human experiments have used environments characterised by a single minimum of error and effort (Fig. 1A). This means that there exists only one solution to the task such that any deviation from this solution would lead to an increase in the magnitude of task error or in the overall effort required from the subject. In the

Challenges from tasks with multiple solutions

The challenge for models to explain these results is two folds. First, the solution choices in a multi-solution environment conform to a regular pattern and cannot be ascribed to noise. For example, in the above perturbation attenuation experiment [22], even after realising a lower energy strategy to perform the task, on being pushed to the high energy strategy, subjects chose to keep it. Although subjects had experienced both strategies with easily distinguishable co-contraction levels, they

Plan as a possible solution

However, it is clear from daily experience that switching between tasks is possible. For example, one can play tennis in the morning, drive to work, work with the different machines in ones office and take a swim after work. So it is possible to switch between different tasks and improve them even if each of these tasks requires coordination between a different set of limbs, is performed with tools of different dynamics and in environments with different properties. So, why can the human CNS

Neural correlates of motion plan and execution

Traditionally, motor control is believed to be hierarchical with sensory information or self-intention leading to a motor action via a planning process. The definition from a popular text book [30] states that motor planning is the translation of a general outline of behaviour to concrete motor response through processing in motor pathways. These pathways have been elucidated over the years from lesion studies, electrophysiology and brain imaging (Fig. 4). Planning probably starts in the

Dissociating motor plan from execution behaviourally

Even though behavioural experiments cannot dissociate planning from execution by the spatial and temporal hierarchy, they can do so by looking for a command hierarchy. As muscle activations yield the final representation of the motor command that drive movements, any feature that is common across muscles would point to a process hierarchically above the muscle level which can be interpreted as a plan [38].

In a recent study [26], we used this idea to gain new insights about motor plans. Subjects

Functional role of plans and memory

The above results suggest that the CNS uses a motion planning phase with multiple plans, and a memory mechanism which acts on the plans as well as at muscle level. This casts a series of questions about the functional role of these mechanisms in motion control and learning.

First, what could be the utility of such a memory mechanism? Real life tasks are characterised by large redundancy, as each task may be solved in many ways by the CNS, both in terms of the plan to be used (trajectory in case

Modelling tasks with multiple solutions

In this article we pointed to challenges faced by optimal feedback control and adaptive control based models of motor control that have been designed based on results from experiments with a single minimum of error and effort. In particular, these models are unable to explain human behaviour in the presence of multiple solutions. Models based on optimal control will tend to predict the (absolute) minimum cost solution in the manifold of solutions fulfilling the task constraints, and thus miss

Motor control strategies from humans to robots

Similar to humans, robots have sensors and actuators that they can utilise to read the environment and make actions. They require to learn how to assimilate the sensory signals into useful information, what actions to perform and how to control these actions — the very questions that drive human motor control research. The study of motor control and learning in humans is thus closely related to robotics as it looks to understand how the human motor system addresses the same questions that

Gowrishankar Ganesh received the B.E. (first-class Hons.) degree from the Delhi College of Engineering, New Delhi, India, in 2002 and the M.E. degree from the National University of Singapore in 2004, both in mechanical engineering, and the Ph.D. degree in bioengineering from Imperial College London, UK, in 2010. He was an Intern Researcher with the Computational Neuroscience Laboratories, Advanced Telecommunication Research Institute, Kyoto, Japan, between 2004 and 2009, where he is currently

References (46)

  • M. Hirashima et al.

    Learning with slight forgetting optimizes sensorimotor transformation in redundant motor systems

    PLoS Computational Biology

    (2012)
  • H.K. Khalil

    Nonlinear Systems

    (2002)
  • E. Todorov et al.

    Optimal feedback control as a theory of motor coordination

    Nature Neuroscience

    (2002)
  • L. Rigoux et al.

    A Model of reward- and effort-based optimal decision making and motor control

    PLoS Computational Biology

    (2012)
  • E. Guigon et al.

    Computational motor control: redundancy and invariance

    Journal of Neurophysiology

    (2007)
  • R. Shadmehr et al.

    Adaptive representation of dynamics during learning of a motor task

    Journal of Neuroscience

    (1994)
  • M.A. Conditt et al.

    The motor system does not learn the dynamics of the arm by rote memorization of past experience

    Journal of Neurophysiology

    (1997)
  • E. Burdet et al.

    The central nervous system stabilizes unstable dynamics by learning optimal impedance

    Nature

    (2001)
  • M. Kawato et al.

    A hierarchical neural-network model for control and learning of voluntary movement

    Biological Cybernetics

    (1987)
  • J.J. Slotine et al.

    Applied Nonlinear Control

    (1991)
  • C. Yang et al.

    Human like adaptation of force and impedance in stable and unstable interactions

    IEEE Transactions on Robotics

    (2011)
  • K.P. Tee et al.

    Concurrent adaptation of force and impedance in the redundant muscle system

    Biological Cybernetics

    (2010)
  • A. Kadiallah et al.

    Generalization in adaptation to stable and unstable dynamics

    PLoS One

    (2012)

    • Cited by (27)

    • A geometric optimal control approach for imitation and generalization of manipulation skills

      2023, Robotics and Autonomous Systems

    • Interception of virtual throws reveals predictive skills based on the visual processing of throwing kinematics

      2022, iScience
      Citation Excerpt :

      Four main classes of throwing strategies were identified across a pool of untrained throwers (n = 20), mainly differing in their stepping pattern and, to a lesser degree, in the trajectory of the throwing arm. The emergence of individual throwing strategies (or styles), above the overall background motor variability characterizing the unconstrained throwing task, finds resonance in the more general view that redundancy in complex motor tasks implies the existence of multiple solutions (Ganesh et al., 2010; Ganesh and Burdet, 2013; Maselli et al., 2019) and with the evidence that individuals tend to adhere to one of the possible solutions (Vidal and Lacquaniti, 2021). These results further motivate and generalize the question above, namely if non-expert throwers can equally extract information from different throwers or are better attuned to a specific class of throwing strategies.

    • The road towards understanding embodied decisions

      2021, Neuroscience and Biobehavioral Reviews
      Citation Excerpt :

      While the two domains of decision-making and motor control have long remained separated, a more recent trend sees motor control as a form of decision-making - or the problem of maximizing the utility of movement outcomes under various sources of uncertainty (Shadmehr et al., 2016; Wolpert and Landy, 2012; Shadmehr and Ahmed, 2020). This analogy licenses the cross-fertilization of ideas and formal tools (e.g., statistical theory) across the two domains; and the development of novel computational models that combine objective functions used in economics, such as the maximization of expected utility, and in motor control, such as the minimization of motor costs (Ganesh and Burdet, 2013; Lepora and Pezzulo, 2015; Wispinski et al., 2018) These formulations become relevant when considering complex redundant tasks for which the same outcome can be achieved with different actions, as in a throwing task for which it is possible to hit a target (a fixed locations) with virtually infinite combinations of the release velocity of the projectile (Fig. 7b). This redundancy poses the issue of selecting one among a set of actions, or sequences of actions, that guarantee the same level of performance.

    • Individuals Prioritize the Reach Straightness and Hand Jerk of a Shared Avatar over Their Own

      2020, iScience
      Citation Excerpt :

      Here, we examined this issue using an arm reaching task in virtual reality, performed by a shared avatar who's arm movement was the average of arm movements by the two participants who share it. Motor neuroscience studies have shown that individual human movement trajectories are a result of sensori-motor planning and control (Wolpert et al., 2011; Franklin and Wolpert, 2011; Ganesh and Burdet, 2013) as well as memory-related processes (Diedrichsen et al., 2010; Ganesh and Burdet, 2013; Ganesh et al., 2010). Considering these results, in this study we investigated what the individuals prioritize when they share a virtual body, specifically whether they still optimize their own movements or whether they optimize the movements of the avatar, and, consequently, how the sharing affects the avatar control.

    • Exploring the relationship between computational frameworks and neuroscience studies for sensorimotor learning and control

      2024, Journal of Autonomous Intelligence

    • Imitation of Manipulation Skills Using Multiple Geometries

      2022, arXiv
    View all citing articles on Scopus

    Gowrishankar Ganesh received the B.E. (first-class Hons.) degree from the Delhi College of Engineering, New Delhi, India, in 2002 and the M.E. degree from the National University of Singapore in 2004, both in mechanical engineering, and the Ph.D. degree in bioengineering from Imperial College London, UK, in 2010. He was an Intern Researcher with the Computational Neuroscience Laboratories, Advanced Telecommunication Research Institute, Kyoto, Japan, between 2004 and 2009, where he is currently a Researcher. Since 2010, he has been a Specialist Researcher with the Advanced ICT group of National Institute of Information and Communications Technology, Tokyo, Japan. His research interests include human motor control, robotics, signal processing, and mechanical design.

    Etienne Burdet received the M.S. degree in mathematics, the M.S. degree in physics, and the Ph.D. degree in robotics from the Swiss Federal Institute of Technology, Zurich, Switzerland, in 1990, 1991, and 1996, respectively. He is currently a Reader with Imperial College London, UK. He is doing research at the interface of robotics and bioengineering and his main interest is in human–machine interaction. He has contributions in various fields from human motor control to virtual-reality-based training systems, assistive devices, and robotics for life sciences.

    View full text
    Copyright © 2012 Elsevier B.V. Published by Elsevier B.V. All rights reserved.