Context
Teleoperation systems with haptic feedback can be used to accomplish tasks requiring human levels of dexterity over long distances. Fields like disaster relief and remote care stand to benefit significantly from systems capable of performing such tasks. However, time delay in communication between operator and telerobot make controlling such systems a difficult task. Traditional approaches involving sending power variables such as velocity and force over communication channels result in stability problems for the system. Methods to mitigate this directly exist, but they trade off stability problems for performance problems. Alternatively, Model-Mediated Teleoperation (MMT) sends information about operator motion to the telerobot, and the telerobot sends a model (or a parametrisation thereof) of the environment back to the operator, which can be used to reconstruct interaction force for haptic feedback. Such systems are stable and perform well provided the model is sufficiently descriptive and it does not change with time.
Impedance Reflection is a form of MMT that uses an estimated parametrisation of the environment’s impedance as a model. Typically, this impedance is a differential equation, and a rendering module needs to integrate this impedance model using a set of initial conditions in order to display a force to the operator.
Problem
An object moving in free space exhibits different dynamics to an object resting on a table, or an object locked in a mechanism. Nevertheless, over the course of a natural interaction with the environment, an object can be reasonably expected to change between any of these states of constraint. The fact that the object dynamics can be different depending on the phase of the interaction needs to be supported in the Impedance Reflection framework, if it is to be used for the manipulation of dynamic objects. This assignment focuses on developing and testing a new design for an impedance reflection system capable of transitioning between different models, using information from previous estimation runs to keep convergence times low, and using information from previous rendering runs to prevent mismatches in transient dynamics between the real and simulated situations, if necessary.
For example, if an operator is manipulating an object, and suddenly swipes it across a table, the existing estimate about mass should be used to augment the estimation of the new dynamics, which involve both the inertia of the object and its friction with the table. Also, the states of the mass (e.g. position and velocity) should be taken into account for rendering force in the new circumstances, used for example as initial conditions in the new rendering run.
Practically, this will involve:
- Investigating the importance of transient dynamics of the environment in standard teleoperation tasks
- Developing an MMT system capable of supporting transition between different models, assuming that the system is already capable of evaluating which model needs to be used
- Validating the system through experiments in a real-world environment