Predictive Avatar Control and Feedback (PACOF)

WP3: Proactive mathematical control in telemanipulation


Funded by:




Chair HS:

Daily Supervisor:



The use of avatar robotics allows human operators to apply their knowledge and skills in situations where their physical presence is not possible, for example due to a hazardous or hard-to-reach environment. Such situations may include both such tasks as the handling of hazardous or toxic substances as well as social and medical care.

Instead of a human a robot, the avatar, takes its place and is used for interactions with an environment. It is remotely controlled by an operator reproducing as good as possible the entered movements and commands, as if the operator itself were in its position. At the same time this operator is isolated from the outside world, instead being provided with realistic stimuli and measurements from the environment around the avatar, experiencing what the robot experiences, again as if the person were physically present.

One of the main challenges in avatar robotics are the inevitable delays that appear in the system due to the information transfer and processing in between the avatar and the operator. Such delays both impede the control experience for the operator and the avatar performance, thus restricting the range of application of such systems.

The idea of the PACOF project is to tackle these problems by predicting both the actions of the operator and the reaction of the avatars environment. This way the avatar would predict what the operator will do next and could already start to execute the operation before the delayed command from the operator arrives. Similarly a model of the environment could predict its response to a particular action the operator just ordered, providing a real time control experience.

Concretely this is achieved by splitting the system into three loops each with a degree of autonomy. A first one on the avatar side predicting the operator input and controlling the avatar, one on the operator side predicting the environment response and providing feedback to the operator and a third in between which connect the other two, continuously updating their models and minimizing mismatches, using the delayed connection between them.

The mathematical part of this interdisciplinary project will then focus on developing novel control techniques for use within these three loops, applying for example energy-based models such as port-Hamiltonian systems and model-free control techniques, such as funnel control.


See article in the from 16-03-2021