Hack the Car by Coffee Driver (Guardian Angel)

Idea for Our Tool: Warning Traffic Participants of Danger Situations using Car2X Communication

The idea for our tool comes from recognizing the need to improve communication between vehicles on the road. Specifically, we aim to address the issue of potential danger situations arising when interacting with other road users. To achieve this, we have developed a tool that uses Car2X communication to warn traffic participants if a danger situation can arise.
By providing clear and timely communication to other road users, we believe we can increase safety and reduce the number of accidents on the road. We recognize the importance of efficient and effective communication and strive to contribute to the development of a safer transportation system. We have chosen to focus on one specific scenario for the SDV Hackathon 2023 challenge due to the limited time frame.

Covered scenario

The scenario we have chosen to focus on for the SDV Hackathon 2023 challenge involves a pedestrian about to cross the street in front of or behind the Ego vehicle while another vehicle (Offender) is passing through. In this situation, there is a risk of an accident occurring, if both the Offender and the pedestrian are not aware of each other's presence. To prevent this from happening and increase road safety, our tool (Guardian Angel) is designed to activate the indicator lights in the Ego vehicle as a warning signal. This warning signal will alert both the Offender and the Victim of the potential danger and allow them to take appropriate actions to avoid an accident. Our tool uses Car2X communication to detect the presence of the Offender and the pedestrian and determine their positions relative to the Ego vehicle. If both are present at the same time, the tool triggers the indicator lights in the Ego vehicle. This feature is designed to provide a clear and timely warning to all parties involved and reduce the risk of accidents.

=======================================



---   ---   ---   ---   ---   ---   ---
  ----------
  |Offender|-------->   !
  ----------
                        ^
    -------   --------- | -------
    |     |   |Ego car| | |     |
    -------   --------- | -------
========================|==============
                        O

Real-life demo video Visualization demo

Herewith is the third traffic participant introduced into usual two-participant traffic situation.

        flowchart LR
            Victim --- car[Ego car] --- Offender --- Victim

Approach : Detection Algorithm and Warning System

Our tool uses the Ego car sensor array to detect two defined traffic participants: the Offender and the Victim. The detection process includes selecting relevant objects and determining their intentions.

The development of the detection algorithm is separated into two complexity stages:

• Stage 1: we use a simple static check to determine if either the vehicle or pedestrian is in the defined danger zone. Implemented.
• Stage 2: we build movement vectors to recognize situations dynamically. Not yet implemented.

If a potential danger situation is identified, our tool activates a warning system. The available Ego vehicle can activate turn signal lights as a warning to all parties involved. In addition to this warning system, other warning channels and visualizations are possible.

• Using car lights to visually alert traffic participants to the warning, specifically highlighting the pedestrian to improve visibility (In-car hardware).
• Car2car communication to warn the Offender via their HMI to prepare for, or execute, braking (Car2car).
• Our tool could utilize the built-in music system to audibly communicate the danger and its direction to the pedestrian (In-car hardware).
• Anonymized post into the (city traffic) Cloud to train the AI and detect potential danger spots (Car2X).
• Activating of pedestrian traffic lights, if such a crosswalk is used by the pedestrian (Car2X).

Realization

Our primary focus in the realization of this tool is to implement the logic for detecting relevant objects and triggering a warning system in case of potential danger situations.

      flowchart LR
          data_source[Sensor data] --> angel[Guardian Angel] --> warning[Warning system]

Both the sensor data and the warning system are provided by the given car over a high-level abstraction layer.

Data flow

        flowchart LR
            classDef given stroke:#777, fill:#777
    
            subgraph "Sensor data"
                Car:::given
                Trace:::given
                Stub
            end
    
            subgraph "Danger detection"
                transformer[Coordinate System Transformer]
                offender[Offender Detector]
                victim[Victim Detector]
                angel[Guardian Angel]
            end
        
            subgraph "Warning rising"
                car_out[Car]:::given
                Car2X
            end
        
            Car & Trace & Stub --Markers --> transformer
            transformer -- Transformed marker--> offender & victim
            offender -- offender detected --> angel
            victim -- victim detected--> angel
            angel -- Light on --> car_out
            angel -- Data --> Car2X

Sensor data feeding

In our development process, we utilize both a recorded trace from a set of sub-scenarios and an implemented stub to populate objects in the Ego vehicle environment. The scenario we are covering involves the detection of two traffic participants: the Offender and the Victim. To better understand the world representation with the provided sensor data, we recorded two isolated traces in addition to a complete trace:

• The movement of the passing car (Offender)
• The movement of the pedestrian (Victim), including movement around the Ego car to determine its dimensions
• Simultaneous movement of both the Offender and the Victim.

These traces provide us with a more comprehensive understanding of the movements and behavior of the traffic participants involved in the scenario. Using this data, we can create an effective detection algorithm that accurately identifies potential danger situations and triggers the appropriate warning system.

Offender and victim detection

The most important functionality of the Guardian Angel is the detection of both the Offender and the Victim. Depending on the parking position of the Ego car a danger zone is placed in the virtual representation of the real world. We defined this zone as a rectangle limited by far and near borders for both longitude and latitude Ego car axles.

The Guardian Angel uses two components to detection of Offender and Victim respectively:

• Offender Detector
• Victim Detector

Both detector modules implemented following functionality:

• Subscribing on the marker message
• Filtering by object type (car/pedestrian)
• filtering by all four danger box borders (coordinates are outside the danger zone)

If all conditions are filled a "detection" flag is raised.

Coordinate system transformation

The trace data provided contains coordinates in a world-related coordinate system. Although the axes of this system may not necessarily be parallel to those of the car-related coordinate system. To detect the parallelism of the Offender trajectory, we require a coordinate system transformation. The offset and angles of rotation for all three coordinates are provided by the car abstraction layer.

This transformation allows us to accurately determine the position and movement of the Offender relative to the Ego vehicle. Through careful analysis and implementation of these transformations, we can create an effective detection algorithm that accurately identifies potential danger situations and triggers the appropriate warning system.

Rising of the warning

The Guardian utilizes the Ego car sensor array to detect potential danger situations involving the Offender and the Victim. If a danger situation is identified, the system triggers the warning system, which includes activating turn signal lights on the Ego vehicle and (for this SDV Hackathon challenge) a simulation of the Car2X

Integration of the Guardian Angel components

Transfer of the data from and to the components is realized with eCAL messages that contains objects in protobuf-format. All components subscribe to certain message topics, which gives the possibility to visualize the inter-component communication and to inject synthetic messages for testing purposes. A stub is developed to facilitate message generation for testing purposes by generating messages with desired values.

Testing

To ensure that our tool functions correctly, it is necessary to test the behavior of each component and the system as a whole. The system has been developed in a modular architecture, with defined data containers (messages). As a result, each component can be tested in isolation, but also as part of a subsystem or the entire system. This approach allows us to better understand and address any issues or errors early on in the development process.

    flowchart LR
        Stub -- Stimulation --> Component -- Result --> Evaluation

    flowchart LR
        Stub -- Stimulation --> Component1 --> Component2 -- Result --> Evaluation

A [test set](Test cases.md) containing a small amount of test cases and a testing component are also available.

Car2X communication channel

For the SDV Hackathon 2023 challenge developed Guardian Angel contains a simple simulation of a Car2X communication. To achieve this target a MQTT broker and a consumer component were integrated (provider and subscriber). Both components establish a communication channel, which can be used to trigger some warning events or messages.

Technology Stack

Here are some of the main tools we used to build our project:

• python 3.11
• eCAL v5.12.1
• Foxglove Studio v1.79.0
• eCAL Foxglove Bridge
• Eclipse Mosquitto (MQTT Broker)