Keywords

1 Introduction

In recent years, autonomous driving has become a focal point of research and development. As a matter of fact, there is currently no automated vehicle that is capable of operating in all conditions and environments, as defined in SAE level 5 [6]. Automated vehicles still require human intervention in unknown driving environments or the event of hardware or software failure. As a transition to fully automated vehicles, teledriving also known as remote operation of vehicles over the internet plays an important role. Teleoperated vehicles can be employed in various practical scenarios, including taxi services, delivery services [2] and shared mobility platforms such as from Vay Technology GmbH [8]. In [7], the structure of a teleoperated vehicle system is described. During vehicle teleoperation, the control signals of the teledriver such as steering input, throttle and brake pedal positions are transmitted to the vehicle control unit (VCU) over the internet and thus the signal transmission is delayed. To provide the teledriver with appropriate visual and acoustic feedback, the teleoperated vehicle has to be equipped with cameras and speakers. Likewise, there is a time delay to reach the telestation. In Fig. 1, the schematic structure of the telestation and the teleoperated vehicle is illustrated.

Fig. 1.
figure 1

Teleoperated vehicle system based on [7]

Full size image

Despite the high potential of the concept of teleoperated vehicles, an investigation into the dynamics and the stability of the vehicle-teledriver system remains absent in the current literature to the best of the author’s knowledge. This paper studies the influence of the communication time delay and the reduced perception of the teledriver on the vehicle-teledriver system. Furthermore, a torque emulation concept is introduced that considers the telestation signals and the vehicle states to improve the teledriver’s haptic perception and the ease of control of the teleoperated vehicle [3].

2 Vehicle-Teledriver System

In this section, the vehicle-teledriver system is derived from the conventional vehicle-driver system. As described in [5], the vehicle-driver system consists of the vehicle model \(G_{Veh}(s)= \frac{y_{CG}(s)}{\delta _{sw}(s)}\), the driver model \(Dr(s)=\frac{\delta _{sw}(s)}{\varDelta y(s)}\) and a prediction transfer function \(Pr(s)=\frac{y(s)}{y_{CG}(s)}\). The steering wheel angle is \(\delta _{sw}\), \(y_{CG}\) is the current lateral position of the vehicle, \(y\) is the predicted lateral position and \(\varDelta y\) is the deviation between the predicted position and the lateral position reference in the local road frame. Firstly, the parameter set of the driver model \(Dr(s)\) is derived for conventional driving based on [5] and [3].

When the driver is physically not present in the vehicle, the communication between the driver and the vehicle takes place via the internet. The teledriver inputs are transmitted to the vehicle with a communication delay. The vehicle’s output and the visual information are transmitted to the driver with a communication delay as well. Therefore, the delay transfer function \(G_{Com}(s)= e^{-T_Cs}\) is added. The resulting closed-loop vehicle-teledriver system is shown in Fig. 2. For the purposes of this investigation, the communication time delay is assumed to be constant \(T_C=0.0\)4 s. The stability of the closed-loop system is evaluated based on the Nyquist stability criterion. The open-loop transfer function is

$$\begin{aligned} G_{o,TS}(s)= Dr(s)\ G_{Com}(s)\ G_{Veh}(s)\ G_{Com}(s)\ Pr(s). \end{aligned}$$
(1)
Fig. 2.
figure 2

Block diagram of the vehicle-teledriver system

Full size image

As shown in Table 1, the phase margin \(\varphi _R\) decreases with increasing time delay until it reaches a negative value. That demonstrates that the closed-loop motion can leave the stable regime due to communication delay. The crossover frequency \(\omega _c={0.285\,\textrm{Hz}}\) does not change with varying communication delay.

Table 1. Influence of the time delay on the system stability
Full size table

Now, it is assumed that the human operator is able to adapt to the situation of driving a vehicle remotely, based on experience. Therefore, it is further assumed that the teledriver is trained to operate the vehicle at a certain speed, e.g. \(v_x=3\)0 km/h, and with a communication delay \(T_C=0.0\)4 s. The parameters of the teledriver model are derived accordingly for this speed. In Table 2, the crossover frequency and phase margin of the closed-loop system are given. As the vehicle velocity is changing, but the teledriver does not adapt due to the reduced perception of speed, it can be seen that the crossover frequency shifts to higher values and the phase margin decreases with increasing velocity, until the phase margin is negative. This results from the non-minimum phase behaviour which becomes more dominant at higher frequencies. The closed-loop motion with the teledriver leaves the stable regime at 45 km/h due to the reduced perception. This phenomenon can also be observed with teledrivers in real application.

Table 2. Influence of the reduced perception on the system stability
Full size table

In conclusion, a teledriver can steer the vehicle in a stable fashion despite the time delay. However, if the teledriver’s perception of the changing speed is reduced, the vehicle motion can become unstable. There might be teledrivers who are able to adapt themselves to different speeds, but the motion could still become unstable due to the communication time delay. To increase the stability of the teleoperated vehicle-teledriver system, a phase-boosting element, such as lead-steering as proposed in [4], could be introduced, which is however beyond the scope of this paper. Alternatively, a steering wheel torque emulation could also be helpful for the driver to better adapt to changing speed.

3 Torque Emulation

In literature, there are many concepts to emulate the feedback torque at the steering wheel. However, most of them require a profound knowledge of the steering system parameters. In the scope of this study, a tunable torque emulation concept is introduced and implemented in a teledriving simulator and a real telestation. Since there are no physical links between the telestation steering system and the vehicle steering system, the feedback torque can be generated freely. The torque emulation concept has a modular structure that can be tuned as described in [3]. The emulated torque considers the telestation steering wheel angle \(\delta _{sw,TS}\), telestation steering wheel rate \(\dot{\delta }_{sw,TS}\), lateral acceleration \(a_y\) and the yaw rate r of the vehicle. A spring torque is calculated based on the steering wheel angle and a damping torque based on the steering wheel rate. The lateral acceleration and the yaw rate based torques are used to provide the teledriver with information of the vehicle motion. Each torque component \(T_i\) is evaluated with a two-dimensional characteristic curve that considers the longitudinal velocity \(v_x\) and the corresponding signal. The characteristic curves are generated with the hyperbolic tangent function as described in [1]. The maximum value and the slope around the origin are varied with the parameters \(A_i(v_x)\) and \(\xi _i(v_x)\). The variable \(\chi \) and the index i describe the individual signal to generate the torque component \((\chi = i=\delta _{sw,TS}\text {, }\dot{\delta }_{sw,TS}\text {, } a_y\text {, } r)\)

$$\begin{aligned} T_i= A_i(v_x)\tanh (\xi _i(v_x)\cdot \chi ). \end{aligned}$$
(2)

For test purposes, a slalom is build up in a teledriving simulator. In Fig. 3, the view of a telestation is shown. The tests at the real telestation can be found in [3].

Fig. 3.
figure 3

Telestation with operator in real environment [8]

Full size image
Fig. 4.
figure 4

Trajectory with different torque emulation modes at \(v_x=3\)0 km/h (left); Steering inputs and vehicle states with different torque emulation modes (right)

Full size image

The reaction of a teledriver with little teledriving experience is tested with different torque contributions. For this study, the results with no torque emulation (mode 0) are compared with two different torque emulation modes. The investigated torque emulation mode 5 only includes the spring and damping torque. Whereas mode 9 considers all four torque components. In the left graph of Fig. 4, the trajectories with the different torque emulation modes at \(v_x=3\)0 km/h are illustrated. In comparison to mode 0, where no feedback torque is applied, the driver successfully navigates through the slalom with the assistance of the feedback torque. Moreover, the amplitudes are reduced with a feedback torque at the steering wheel. However, in mode 5 the driver leaves the track after passing the exit gate of the slalom. In contrast, in mode 9 the driver finishes the slalom without touching any cones and not coming off the track. Additionally, the driver manages to steer the vehicle with a slightly smaller amplitude than in mode 5. The steering inputs of the driver and the corresponding vehicle states with different torque emulation modes are compared in the right graph of Fig. 4. With the feedback torque at the steering wheel, the driver steers less than without a feedback torque. Furthermore, in mode 9 the driver reacts earlier passing the cones than in mode 5 or mode 0. Additionally, the steering wheel angle rate is lower in torque emulation mode 5 and 9 than with mode 0. Due to the lower steering wheel angle input in modes 5 and 9, the lateral acceleration and the yaw rate are also lower. The tuning parameters of mode 5 are chosen such that the spring torque has the greatest contribution. This steering wheel angle based torque may give the driver the feeling of having an immediate response, but the information of the vehicle’s response is not provided. In mode 9 the contribution of the spring torque is smaller, but there are additional torque components such as the lateral acceleration and the yaw rate based torque. The torque emulation component was also introduced to a real telestation. The tests in the real system show that torque emulation assists the steering of a teleoperated vehicle and improves the teledriver’s perception of speed and the road, which is helpful when performing the slalom manoeuvre from simulation in a real environment [3].

4 Conclusion

This study shows from a theoretical perspective how the teleoperation of a vehicle differs from driving a normal vehicle. In general, the teledriver has to cope with the influences of the varying time delay and the reduced perception. The combination of these two impacts makes the teleoperation of a vehicle more challenging than conventional driving. To increase the driver’s perception and the ease of control of a vehicle, a basic and tunable torque emulation concept is introduced, that considers the steering wheel angle, steering wheel rate, lateral acceleration and the yaw rate. Due to the simple structure, it is possible to run the torque emulation component in real-time on a simulator and the real telestation.

A teledriving simulator for testing the driving behaviour in a safe environment is used. The tests at the teledriving simulator demonstrate that a non trained teledriver may have difficulties driving without any torque feedback. Generally, the absence of feedback torque gives drivers an unusual light feeling. That leads to heavy steering due to the reduced perception. It could be demonstrated that the motion gets unstable whenever big steering wheel angle inputs are applied in combination with a high steering wheel angle rate. In this case, the driver tries to steer more due to the delayed reaction of the vehicle. Furthermore, it became clear that it is beneficiary to have a component in the teleoperated vehicle-teledriver system that boosts the phase of the open-loop system and therefore ensures the stability of the closed-loop system. For further investigations the torque emulation component could be integrated into the system model to investigate its influence on stability.