Leveraging Human Intent for Shared Autonomy
As vehicles become more autonomous, it is imperative that mutual transition of control between the vehicle and driver be both safe and smooth.
Considering the progression of vehicle autonomy, levels 3 and 4 (shown below) have both autonomous and human-driver modes of control.
These levels of autonomy allow for a form of collaborative robotics. To frame this, consider the time leading up to a possible collision:
The following areas immediately lend themselves to collaborative robotics considerations:
Within this region, transitioning control authority between the human and vehicle is an important consideration.
The question is: "When is control transition allowable"? If control can be transitioned at a non-zero speed, then this handover should be safe.
When control is abrupt, the human take-over control policy may or may not remain in a 'safe-set'. An easy example is maintaining a lane:
If the vehicle is able to model through behavioral modeling (cloning), safe human control, then current human control input can be compared to the expected safe control. If the human's control input is contained in the safe control input set, then the vehicle can smoothly transition control over to the human driver:
To implement this, the vehicle must be capable of modeling a) its own dynamics b) the human driver behavioral model c) the external environment. This is represented below:
In addition to strict observation of the human driver, the vehicle may also present risk (well before mandatory take-over is required), as well as display it's own perception (which may inform the driver if there is a possible fault):
The expected outcome is a control paradigm where control-authority transition between the vehicle and driver can be managed through behavioral cloning and risk presentation