Functional requirements

In order to shed ligh on the capabilities and functions that are expected from the developing system which it must do, the functional requirements are created and

56

described. They help to determine what should be implemented in order to meet the goals for this work from functional perspective.

Its functional model using IDEF0 standard has been established to evaluate the functional specifications for the information system. A functional modeling methodology and graphical notation designed to formalize and explain business processes is the IDEF0 standard. (Tsiring and Beker, 2020) A characteristic feature of IDEF0 is its emphasis on subject subordination. IDEF0 takes into account the logical relations between works, not their temporal sequence. (Tsiring and Beker, 2020)

The IDEF0 standard represents the organization as a set of modules, here the input arrow always comes to the left edge of activity, the control arrow - to the top side.

output - right side.

The definition looks like a "black box" with inputs , outputs, power, and process that is increasingly detailed to the required level. (V. Moroz and V. Baulina, 2018) In addition, there are dictionaries explaining the actions and arrows to be correctly interpreted. (V. Moroz and V. Baulina, 2018) You may explain in these dictionaries what material you put into the operation or arrow. (V. Moroz and V. Baulina, 2018)

Using the IDEF0 standard, all main functionalities are described, so that, they can be developed correctly, according to the work objectives.

Figure 2 presents a conceptual diagram that reflects the view of the functions of the system.

Figure 3 – A conceptual diagram of the model

57

The conceptual diagram depicts the main function of the system, which is the long-term prediction of surrounding agents motion in urban environment, its inputs, such as information which comes from the sensors of the vehicle, and history of surrounding agents motion. On the top side there are control signals, which are traffic rules and maps of the environment. The output of the function is the plan for vehicle’s motion.

Figure 3 represents a decomposition of the conceptual diagram with its main functions.

Figure 4 – A decomposition of the conceptual diagram that represents the main functions of the model

A decomposition of the conceptual diagram depicts main functions which make up the main function. As it can be seen, at its left side, there are two functions, the Object Detection function and Sequence to sequence analysis of agent’s motions.

Speaking about Object Detection, its purpose is to detect objects and send an appropriate information to the next function, the Object Tracking. Object Tracking, in turn, generates a tracking motion and sends the output data to the Prediction module.

As it was discovered during the interviews, “Since 2011 it has been the rise of neural networks and in computer vision they have outperformed classical methods.”

(Autonomous Driving Researcher, personal interview, 2020) So that, the object

58

detection and object tracking modules are taking ideas from the approaches based on latest achievements in the area of computer vision and neural networks as the most relevant and robust for the developing model.

At the same time, the Sequence to sequence module generates motion sequences information based on the agent's information and sends it to the Prediction.

During the investigation of problematic of prediction of surrounding agents motion in urban environment, as it was mentioned in appropriate section of this work, a long-term prediction method is highly dependent on the motion history of the agent, so it is necessary to have a robust and reliable motion history in order to get the most efficient and safe predictions.

Thus it is essential to have a robust and efficient tool to generate historical knowledge and based on that knowledge produce efficient sequence of predictions and decisions during the future motion process of the autonomous vehicle.

The Prediction, in turn, based on the input data generates its predictions, then, this information, combined with information about ego-vehicle position is going to the Planning which purpose is to generate the motion plan.

It was discovered in the beginning of the work, that it is essential for the model to be able to perform the path planning and eliminating obstacles is about finding the shortest and fastest routes to a destination. Route planning begins with a long-range strategy when you put an address into a map application. (How Do Self-Driving Cars Work? | The Zebra, 2019) Then, as the car drives, short-range plans are created and continuous refinements are made. (How Do Self-Driving Cars Work? | The Zebra, 2019)

Figure 4 represents a decomposition of the function Prediction.

59

Figure 5 – A decomposition of the function “Prediction”

A decomposition of the Prediction function depicts its main sub-functions, evaluation of the tracking history and motion sequences, the flow-based prediction and the trajectory particle filter.

As it is stated in the system requirements, The model should be able to contemplate and predict the movements of the surrounding agents recognized on one of the previous steps. It should be able to predict and return trajectories, which include velocity, heading and positions of all tracked agent, that surround the ego-vehicle for several seconds in the future.

As inspiration for the prediction model a long-term prediction approach based on a combined definition of the trajectories and a particle filter structure, suggested by Hermes and Wöhler in their work “Long-term Vehicle Motion Prediction” was used.

(Hermes, Wöhler, et al., 2018). As a measure for similarities between trajectories, the quaternion-based rotationally invariant longest common subsequence (QRLCS) metric used. (Hermes, Wöhler, et al., 2018)

Thus, the functional modeling allowed clarifying the functional requirements for the developed system.

60

Functional requirements for the model of long-term prediction of surrounding agents motion in urban environment:

1) The model has to be able to perform the object detection 2) The model has to be able to perform the object tracking

3) The model has to able to perform sequence to sequence learning of the upcoming sequences of data

4) The model has to be able to perform prediction of agents’ trajectories for several seconds in the future

5) The model has to be able to perform localization of the ego-vehicle position 6) The model has to be able to perform a planning for the efficient motion

61

7 Design a model based on conducted analyses