Safety Performance Assessment via Virtual Simulation of V2X Warning Triggers to Cyclists with Models Created from Real-World Testing

By inergency On Jan 10, 2024

2.1. V2X System Technology and Physical Testing Setup

A prototype on-board unit (OBU) device with, among other features, V2X communication capabilities, was designed and installed on a commercial electric cycle (Kona Dew-E manufactured by Kona Bicycles, Ferndale, WA, USA). This OBU incorporated all the necessary hardware and software modules to facilitate ITS-G5 V2X communications, including the transmission and reception of ETSI standardized CAM messages. The device’s precise location is determined using a high-performance positioning engine with a multi-band Global Navigation Satellite System (GNSS) receiver, in combination with an inertial measurement unit that, in addition to its own accelerometer, gyroscope, and magnetometer sensors, can also utilize the GNSS receiver signals to provide enhanced positioning results via its own fusion engine.

The cycle had some pre-installed hardware features provided by the manufacturer. It came equipped with a common bike computer that accumulates real-time information about the bike’s status, including current speed (via a Hall sensor installed on the back wheel), current cadence (measured via the electric motor’s controller), battery status, and more. This bike computer was connected with the OBU device via BLE (Bluetooth Low Energy), making all the cycle’s real-time information available to the OBU. To obtain the brake status of the cycle, which was not initially available, the team installed Hall sensors in both the front and back brake levers. These brake sensors were directly connected (via a wired connection) to the OBU, enabling real-time monitoring of the brake status. In addition to enhancing awareness, the brake status was essentially used as a trigger for measuring cyclists’ reactions after an emitted warning.

Furthermore, the cycle was also equipped with a touch LCD display, where the cyclist was informed about sensor outputs (positioning, speed, heading, battery status). Using this human–machine interaction (HMI) element, the user could also interact with the device, by starting and stopping on-demand its operation, for better control over the experiments. There were also visuals on the screen that presented collision warnings (Figure 1) and were accompanied with an acoustic buzzer sound in the case of a collision detection, so that the cyclist could become aware of potentially dangerous situations as fast as possible after the detection from the system, without having to keep his gaze constantly on the screen, in order to take avoiding action.

The key features of the OBU enhanced cycle are summarized below:

Short range communications with V2X ITS-G5;
Localization via multiband GNSS receiver fused with IMU;
Real-time monitoring of speed, brake and cadence;
Basic visual and acoustic interaction with the rider via LCD display and a buzzer respectively.

To fully exploit the enhancements on the cycle and evaluate the actual performance of the developed system, realistic riding behavior from a real cyclist on a cycle, who could react to warnings as a human would, was deemed necessary. The physical testing was conducted with 12 volunteers at an IDIADA proving ground in Santa Oliva, Spain. The testing scenarios were based on the findings of the accidentology study conducted in [1] and characterized within [16] to fit Euro NCAP testing protocols and physical testing requirements. For safety reasons, using a real vehicle in scenarios with a high probability of collision between the vehicle and a real cyclist was not feasible. Therefore, a virtual vehicle was used instead. The term “virtual vehicle” refers to a V2X device station that transmitted a pre-recorded series of CAM messages, precisely corresponding to the vehicle’s path, speed, and direction for the scenario being tested in each run. With this strategy, in the V2X ecosystem, the vehicle was “present” without posing any collision safety risks during the experiments. The volunteers riding the cycle were instructed to follow predefined paths and speeds that corresponded to the chosen scenario being tested each time. Figure 2 shows the setup of the physical testing environment.

The selected metric for triggering the warning was time-to-collision (TTC). A threshold of four (4) seconds was adapted as the TTC that triggers a “lighter” cyclist warning, corresponding to an orange indication on the LCD display and a threshold of two (2) seconds for a “stronger” warning, corresponding to red indication on the screen. In both warning cases, however, the buzzer sound was also triggered. A typical V2X application latency timing is the time period between the “birth” of data from sensors of a transmitting ITS station (time at which the transmitted sensor data are available), until this information is received, processed, and interpreted into a meaningful user warning at the receiving ITS station and must be less than 300 milliseconds in case of transport safety applications [18,19,20]. This latency is highly affected by the “freshness” of sensor data, station’s hardware capabilities and computational resources, software and application algorithm efficiency regarding processing time, and finally HMI response time. The minimum TTC needs also to consider the user (in our case, rider) reaction time, containing the human perception and interpretation of the HMI information, as well as the human action upon the emitted warning, the time needed by the bicycle to physically perform the rider’s intentions (brake the bicycle), plus finally an error margin compensating mainly for possible positioning inaccuracies. The 2 s stronger warning (~1.7 s for user perception and action based on the above explanation) was selected as the minimum time that could enable collision avoidance and is consistent, for example, with Euro NCAP recommendations [21] where points are awarded when a forward collision warning is issued at a TTC equal or greater than 1.7 s. Furthermore, previous research by the authors [22] about rider reaction times in a motorcycle simulator after HMI stimulus provided similar results. The lighter warning of 4 s was selected in order to compensate for user unfamiliarity with the experiment scene and the enhanced bike itself. Especially, the installed V2X antennas in the steering could make even an experienced rider a little uncomfortable at first. In a real-world road environment, a 4 s TTC warning will probably lead to alerts even in cases where the rider is perfectly aware and in control of the situation. On the other hand, it will also enable a more comfortable reaction and manoeuvre from the recipient of the warning.

2.4. Evaluation Scope and Research Question

The evaluation scope was the potential benefit assessment of a V2X-capable cycle on-board unit triggering a warning signal to the cyclist in a potential collision scenario, which represented crossing right, crossing left, and car turning left with cyclist in same direction scenes (Figure 3). These scenes have been defined in “Safe-Up” project [1] since they represent a majority of crash scenes in accident databases and additionally are not yet fully covered by EuroNCAP.

The following slightly modified research question was finally derived for the assessment in “Safe-Up” project [16]:

“What is the safety benefit of a VRU C-ITS warning system on connected cyclists in supporting them to mitigate safety-critical events with passenger cars, triggered by a radio signal based (OBU, VRU-smart device) communication and detection system, in terms of accident avoidance compared to Car to VRU collisions on urban roads?”

The idea behind this research question is a quantitative comparison of accidents in urban roads that happened over the last years in German cities to a potential integration of a cyclist communication device, which triggers a warning signal to a cyclist.

2.6. Cyclist Model Generation

The safety benefit assessment is investigating the ability of a cyclist to prevent a collision when receiving a warning signal. To assess such kinds of behavior, a model is needed in simulation which represents a typical cyclist behavior. A wide range of behavioral models can be used in simulation, such as mathematical models describing the trajectory and decisions made of an agent [26] and more sophisticated models trying to model human cognition and interactions with the environment [27]. In Safe-Up, physical tests with real cyclists were performed in three different scenes, as depicted in Figure 3, to assess the reaction of cyclists to a warning signal. The measured trajectories were used to create a mathematical model for typical reaction times to a warning trigger at t = 0 s and the resulting brake distance and acceleration. Due to the scenario-based approach of the available GIDAS scenarios and the event-based triggering of the cyclist, this kind of model was seen to be sufficient for the performance assessment in this study.

Figure 5 shows all cyclists’ velocities over time in the three assessed scenarios. The reaction was similar in all scenes to the warning trigger. The horizontal part of the curves represents the reaction time to the warning trigger, whereas the pitched part represents the continuous braking and thus constant deceleration of the cyclist to a standstill.

Assuming a normal distribution for each of the two segments of the curve (reaction time and acceleration), the behavioral distribution can be computed and assessed. A 90% probability range coverage leads to the following min. and max. distribution values (see Table 1), which have been used as cyclist model parameters. The lower bound of this range (high negative acceleration and small reaction time) is classified as the “progressive” driver, which reacts early to a warning trigger and brakes hard. As an opposite, the upper bound is classified as the “defensive” driver, someone who takes longer to react to a warning trigger and decelerates less progressively. These two types of models may not represent individual driver reactions anymore but can be used to assess the majority of cyclists’ behaviors and the effect of safety technologies in simulation. Thus, a reasonable range of driver behavior characteristics can be shown and tested in simulation.

Based on these results, a functional mock-up unit (FMU) can be created, which is then deployed in the virtual simulation, see Figure 6. The unit works the way that the initial path of the cyclist is followed in the beginning of the simulation. As soon as the warning trigger is sent, the cyclist will react and overrule the given trajectory with the defined behavior.