A cortical network for control of voluntary movements in a robot finger

Abstract

Neurobiological control systems have long been studied as possible inspiration for the construction of robotic controllers. In this paper, a control of voluntary joint movements using a cortical network within constraints from neurophysiology and motor psychophysics is presented. The neural controller is proposed to control desired joint trajectories of an anthropomorphic robot finger. Each joint is driven by an agonist–antagonist actuator pair with muscle-like properties. Dynamical neural network proposes functional roles for pre-central cortical cell types in the computation of a descending command to spinal alpha and gamma motor neurons. Neurons in anterior area 5 are proposed to compute the position of the link in question using corollary discharges from area 4 and feedback from muscles spindles. Neurons in posterior area 5 use the position perception signal and desired position signal to compute a desired joint movement direction. The network reinterprets the “neural population activity” to afford unified control of posture, movement and force production. In addition, it suggests how the brain may set automatic and volitional gating mechanisms to vary the balance of static and dynamic feedback information to guide the movement command and to compensate for external forces. The reliability of the cortical neural controller is demonstrated by experimental results, where control system exhibits key kinematic properties of human movement, dynamic compensation and includes asymmetric bell-shaped velocity profiles.

Introduction

Modern robotics research is concerned with the control of complex plants. Such plants exhibit no-trivial dynamics and potentially long feedback delays. However, in order to be successful, many control techniques require accurate models of both the plant and the environment within which the plant interacts. In traditional form, the control of robots can be stably controlled due to the fact that each joint can be treated as an independent entity, stable high-speed control of a general dexterous manipulator by using tendons (performance redundancy) or more strongly interdependent joints is, to date, highly problematic. In biology, both actuator and plant are in a constant state of flux, and behave in a complex and non-linear fashion. Delays of the sensory–motor loops are typically measured in tens or hundreds of milliseconds. Millions of years of evolution have developed biological controllers, which are very good at controlling these systems. These controllers do not rely on high-quality, pre-defined models of the plant. Instead of this, the control algorithm is tuned incrementally through experience with the environment. It is not surprising, then that we should turn to biology for inspiration.

With this goal in mind, the interface between biology and robotics is carried out by biorobotic researches. Biorobotics tries to emulate the very properties that allow humans to be successful. Each component of a biorobotic system must incorporate the knowledge of areas as diverse as neuromuscular physiology, biomechanics, and neuroscience to name a few, into the design of sensors, actuators, circuits, processors, and control algorithms (see Fig. 1).

In recent years, neurophysiological experiments have made progress in characterizing the role played by various neural cell types in the central nervous system (CNS) for the control of the voluntary natural movements such as reach and grasp [27], [36]. The computational study of motor control is fundamentally concerned with the relationship between sensory signals and motor commands. It is now clear that such movements are the results of a distributed control process, involving a complex array of sensorimotor structures.

At the cortical level, a number of different areas are connected in the planning and execution of such movements [42]. Empirical research on the control of primate reaching movements has ranged from the studies of muscle activity through recordings from cells in the cerebral cortex of monkeys performing reaching tasks to observations of human movements in unusual force environments. In this way, neurophysiological experiments have addressed issues such as the coordinate frames used in motor cortex and post-central areas [1], [27], the relation of cell activity to movement variables [56], preparatory activity [1], load sensitivity [39], the latencies of various responses and equilibrium point control [13]. The voluntary movement system involves four basic functions implicated in the primary motor cortex (area 4) and parietal cortex (especially area 5): (a) continuous trajectory formation; (b) priming, gating and scaling of movement commands; (c) static and inertial load compensation; and (d) proprioception. Early studies during which single-joints were used demonstrated that the discharge of many motor cortical cells co-varied with kinetic parameters of movement, including force, torque, and muscle activation levels [12], [19], [34], [61].

Activity interpretable as motor command priming has been observed in areas 5 and 4 [1], [16], [56], and continuous, scaleable and interruptible activities corresponding to evolving trajectory commands have been observed in area 4 [59]. Voluntary forelimb activity in primates is specialized for transporting and manipulating a wide range of objects of diverse mass. Controlling such movements requires accurate proprioception despite load variations, as well as finely graded force generation, in order to compensate for both inertial and static loads associated with the manipulated objects. Activity interpretable as static and inertial load compensation has long been associated with area 4 [21], [42], and a proprioceptive role for area 5 is also well established [56]. The discharge of many motor cortical cells is strongly influenced by attributes of movement related to the geometry and mechanics of the arm or limb and not only by spatial attributes of the hand trajectory [60].

Toward this goal, as part of an attempt to unify these diverse experimental data, Bullock et al. [5] proposed a computational model that incorporates model neurons corresponding to identified cortical cell types in a circuit that reflects known anatomical connectivity. Furthermore, under computational simulations Bullock et al. showed that properties of model elements correspond to the dynamic properties of many known cell types in areas 4 and 5 of the cerebral cortex. Among these properties are delay period activation, response profiles during movement, kinematic and kinetic sensitivities, and latency of activity onset [18], [39], [42], [47], [58], [60].

In this work, a control of voluntary joint movements for an anthropomorphic robot finger is presented. The cortical level neural controller is based in a biologically inspired neural model presented in [25]. The neural circuit applies a strategy of trajectory control using an extension and revision of the vector integration to endpoint (VITE) model [6], [13], which exhibits key kinematic properties of human movements, including asymmetric bell-shaped velocity profiles. In the cortical network, control signals are not in form of torques to be applied to the joints, but instead control is performed directly in muscle space.

The goals of this research work are (1) to apply knowledge of human neuro-musculo-skeletal motion control to a biomechanically designed and neural controlled robotic system, and (2) to demonstrate that such a system is able to obtain responses in a similar way to the voluntary human joint movements in comparable experiments. This work is a step in the direction toward understanding the working and possibly, the capabilities of the neural circuits in controlling highly-non-linear systems and the structure of the CNS in living systems.

In particular, the experimental platform is a two degree-of-freedom (DoF) robot finger. The actuation system of the robot finger consists of four artificial muscles and their tendons. Each joint is actuated by a pair of artificial muscles (included the tendons). Furthermore, each pair of muscles pulls against one another, which means that the muscles only pull but not push.

This paper is organized as follows. Firstly, the neural controller of trajectory generation for voluntary joint movements of a biomechanically-designed robotic system is described in Section 2. The implementation of the cortical neural circuits on a digital signal processor (DSP) is presented in Section 3. Experimental results with the proposed scheme for control of voluntary reaching joint movements on an anthropomorphic finger are addressed in Section 4. Finally, a discussion on the neural network based in experimental results and future works are given in Section 5.