The main goal of TRACTAT is to lay the foundation for a smooth and effective Transfer of Control (ToC) between autonomous systems and humans within cyber-physical environments.
TRACTAT addresses the following scientific challenges:
- How can the decision to hand over control from an autonomous system to a human be integrated into a system’s planning algorithm in order to find the right point in time and the right target agent for the switchover, as well as estimate constraints for the time frame, during which the interaction needs to be completed? How can the consequences of a planned ToC be estimated?
- How can an autonomous system, in case of an unexpected situation, transfer control to a human employing all available means of multimodal output/presentation, thereby informing the user-with appropriate complexity - about the current situation (situation summary), the system’s previous plan (partial plan presentation), and the cause for the switchover (explanation)? What is the fastest, the most efficient, the most reliable, the safest way to do so when taking into account the user’s attention and cognitive state?
- Likewise, how can a human, in a demanding situation, hand over control of a task to an autonomous system (fast teach-in) employing all available means of multimodal input, thereby informing the system about the current situation and plan? How should the system efficiently request missing information?
- How should an autonomous system deal with situations in which a human forcefully takes over control? How can a system proactively request control from a human?
- Which means of multimodal interaction are suitable to return control to the previous agent after an intermediate takeover of control is complete?
- Which dialogue management techniques can improve the ability to perform the ToC between human and system, such as backchanneling and temporal references, in case the process triggers a dialogue?
- Which parts of the world model have to be exchanged in order for an autonomous system to transfer control over a task to another system with different sensors and actuators, and how can conflicts regarding different views of the world state be resolved?
Application Domains and Scenarios
In investigating the relevance of different application domains for the ToC concepts developed in TRACTAT, the space of possible applications was analyzed. There are two criteria that characterize the complexity of ToC:
- the complexity of the situation, where more context-involving situations require a more elaborate machine-to-human transfer with more explanation, and
- the complexity of the task, where more sophisticated tasks require a more extensive set of interactions as part of the human-to-machine transfer.
A traditional production robot that applies a fixed, trained motion in a secluded environment represents a simple case w.r.t. both situation and task, and hence ToC. Opposed to these straightforward use cases, there are also more unclear or “gray” situations with higher complexity. These are the applications that are the most interesting for TRACTAT
The TRACTAT project investigates the following scenarios in more detail:
- Industry 4.0
- Autonmous Driving