A relational representation for procedural task knowledge
AAAI'05 Proceedings of the 20th national conference on Artificial intelligence - Volume 3
Learning to coordinate controllers-reinforcement learning on a control basis
IJCAI'97 Proceedings of the Fifteenth international joint conference on Artifical intelligence - Volume 2
Sensory integration with articulated motion on a humanoid robot
Applied Bionics and Biomechanics
| Hi-index | 0.00 |
The achievement of a wide variety of tasks using a complex system in an unknown environment presents formidable challenges to the control system and its designer. This paper presents a hybrid DEDS approach tothe control of such systems which allows for reactivity in the continuous domain and for the automatic generation of the control strategy in the discrete framework, thus drastically reducing the amount of system specification required from the designer. In this framework, control is constructed on-line by activating convergent, reactive controllers in a task dependent fashion. Using certain properties of these control modules and a predicate space characterization of their behavior in terms of their control objectives allows thereby to derive ``safe'''' activation strategies automatically using formal techniques from the DEDS formalism. This, the general, largely device independent character of the underlying control modules, and the absence of a requirement for a monolithic control law then drastically reduces the complexity of the control task and thus allows the application of this approach to a large variety of complex task domains. In addition the resulting supervisory control structure can be used as a basis for on-line adaptation and learning mechanisms due to the inherent compression of the state space and the enforcement of ``safety'''' constraints.