Paradoxes in haptic sensing

Citation

Townsend, W. (2004), "Paradoxes in haptic sensing", Sensor Review, Vol. 24 No. 1. https://doi.org/10.1108/sr.2004.08724aaa.002

Paradoxes in haptic sensing

Paradoxes in haptic sensing

William TownsendPresident and CEO, Barrett Technology, Inc., 139 Main Street, Kendall Square, Cambridge, MA 02142-1528, USA. Tel: +617-252-9000; Fax: +617-252-9021; E-mail: wt@barrett.com

Keywords: Haptics, Force sensing, Noncolocation, Robots, Current control

There are two paradoxes in haptic (force) sensing that can be presented as a bad news- good news set. First the bad news: it is extraordinarily difficult, if not impossible, to implement robustly stable sensor-based force control due to a phenomenon known as sensor noncolocation. Noncolocation, in which significant dynamics separate the motor and sensor in the feedforward part of a control loop, is unique to force-control and has no parallel in conventional trajectory-controlled robots. The second paradox and contravening good news: force can be controlled robustly without sensing force.

Haptics enables users, with the aid of a computer-driven haptic device, to interact physically with the geometry of a computer- simulated virtual object by feeling its inertia, softness, texture, slipperiness, and shape. The sole job of the haptic device is to communicate complex worlds through the forces and torques of contact that it generates between itself and its user's hand or other user body part. Conceptually, the collection of all applied forces and torques can be called a generalized contact “force” between the haptic device and the user.

Haptic devices are constructed of mechanical linkages driven by motors. A small haptic device may resemble a joystick or stylus with motors and mechanisms, while larger haptic devices resemble robotic arms. But robot designs optimize computer control of trajectory – positions, velocities, and accelerations. In haptic devices, even though position, velocity, and acceleration are often factored into the computer calculation of force, the trajectory is imposed by the user and not controlled by the computer – the opposite of robots. This functional difference between robots and haptic devices is at the root of the sensor noncolocation issue.

Noncolocation is independent of the sensor quality or bandwidth. While low-dynamic-range touch-pad sensors frequently fail when applied to uneven surfaces, high-dynamic-range industrial strain-gauge-based and optical-based force/torque sensors are robust, precise, and responsive. Indeed, a high-fidelity sensor integrated thoughtfully at the user-end of a haptic device, structurally adjacent to the user contact point, enables precise knowledge of contact force in realtime.

In conventional robotics, where zero- backlash is a safe assumption, even at the expense of high friction, the position, speed, and acceleration of all parts of the system from the motors through the linkages move in strict lock-step according to the kinematic equations. Kinematic motions are fundamentally independent of dynamics. As movement occurs at a motor, the corresponding link motions occur simultaneously. Without dynamic input- output lag between the motor and the links, the sensors are said to be dynamically collocated with the motors regardless of the geometric separation. The Bode plot for this relationship is a flat line and the phase lag is zero at all frequencies.

Now take the case of a haptic device. Torques from the motors, conceptually the motor “force,” ultimately produce the user contact force. In this situation, the inertias and masses of the motors, mechanical transmissions, and links combine conceptually into a generalized “Mass,” intervening dynamically between motor force and contact force. This intervention is called noncolocation even if the motor and sensor are geometrically proximate. The Mass in the contact force to motor force transform equation is a second-order filter in the frequency domain with a 40dB/decade fall-off when graphed on a Bode plot. In real systems, this second-order mass filter easily pushes the system into instability with a limit cycle that is the characteristic of contact force-sensing instabilities. Reducing rate of change in the system by reducing the control gain or adding contact compliance tends to stabilize the system, if not robustly. Since the bandwidth of the forward dynamics are usually of the order of hertz and people can sense forces of the order of hundreds of hertz, the loss in responsiveness severely limits the performance.

Fortunately, there is a method to apply forces and torques with both precision and robust stability. Eliminate the force sensor and then set the torque applied to the motor rotor through precise control of the motor-winding currents. There is a monotonic relationship between motor current and motor torque. Some pundits claim that force control is impossible without a force sensor from which to control force. But as long as the dynamics are made reasonably invisible – fast dynamics and low friction – then, like the force you apply through chop sticks, you can be confident of precise and responsive control of the contact force.