Page 78 - Data Science Algorithms in a Week

62 Olmer Garcia and Cesar Diaz

field (Siegwart, Nourbakhsh, & Scaramuzza, 2011). The navigation field organizes its
techniques into two groups: planning and reacting. The techniques from the planning
group are known as global path planning and are concerned with the generation of the
global route that guides the vehicle toward a goal position. The techniques from the
reacting group are known as local path planning and are concerned with the generation of
several local paths that allow the vehicle to avoid obstacles. In this layer, machine
learning techniques are used to select routes (global and local).
Finally, the control layer will manipulate the degrees of freedom of the autonomous
vehicle (e.g., steering, braking, gearbox, acceleration) for bringing it to the desired
position at a defined speed at each instant of time. Machine learning techniques have
been used to obtain mathematical models and/or adapt a controller to different situations.

Figure 4. Interactions of the proposed cooperative strategy with the architecture of the autonomous
vehicle VILMA01 (Bedoya, 2016).

This research studies the architecting of the layers using a cooperative strategy based
on risk analysis. The resulting architecture includes mechanisms to interact with the
driver (this architecture has been proposed in VILMA01 - First Intelligent Vehicle of the
Autonomous Mobility Laboratory). We stated above that the motion control layer is the
one in charge of manipulating the degrees of freedom of the car (steering, braking, and
acceleration). This manipulation will bring the autonomous vehicle to the desired position
at each point in time. We will explain that this can be achieved by using a predictive
control technique that relies on dynamic models of the vehicle to control the steering
system. The path-planning layer will have the reactive part also known as local path
planning, where the desired path is represented in a curvilinear space. The desired path is