This paper presents our research platform SafeVRU for the interaction of self-driving vehicles with Vulnerable Road Users (VRUs, i.e., pedestrians and cyclists). The paper details the design (implemented with a modular structure within ROS) of the full stack of vehicle localization, environment perception, motion planning, and control, with emphasis on the environment perception and planning modules. The environment perception detects the VRUs using a stereo camera and predicts their paths with Dynamic Bayesian Networks (DBNs), which can account for switching dynamics. The motion planner is based on model predictive contouring control (MPCC) and takes into account vehicle dynamics, control objectives (e.g., desired speed), and perceived environment (i.e., the predicted VRU paths with behavioral uncertainties) over a certain time horizon. We present simulation and real-world results to illustrate the ability of our vehicle to plan and execute collision-free trajectories in the presence of VRUs. I. INTRODUCTION Every year between 20 and 50 million people are involved in road accidents, mostly caused by human errors [1]. According to [1], approximately 1.3 million people lost their life in these accidents. Half of the victims are vulnerable road users (VRUs), such as pedestrians and cyclists. Self-driving vehicles can help reduce these fatalities [2]. Active safety features, such as autonomous emergency braking (AEB), are increasingly found on-board vehicles on the market to improve VRUs' safety (see [3] for a recent overview). In addition, some vehicles already automate steering functionality (e.g., [4], [5]), but still require the driver to initiate the maneuver. Major challenges must be addressed to ensure safety and performance while driving in complex urban environments [6], where VRUs are present. The self-driving vehicle should be aware of the presence of the VRUs and be able to infer their intentions to plan its path accordingly to avoid collisions. In this respect, motion planning methods are required to provide safe (collision-free) and systemcompliant performance in complex environments with static and moving obstacles (refer to [7], [8] for an overview). In real-world applications, the information on the pose (i.e., position and orientation) of other traffic participants comes from a perception module. The perception module should provide to the planner information not only concerning the current position of the other road users, but also † The authors equally contributed to the paper.

Driving simulators are widely used for understanding human–machine interaction, driver behavior and in driver training. The effectiveness of simulators in this process depends largely on their ability to generate realistic motion cues. Though the conventional filter-based motion-cueing strategies have provided reasonable results, these methods suffer from poor workspace management. To address this issue, linear MPC-based strategies have been applied in the past. However, since the kinematics of the motion platform itself is nonlinear and the required motion varies with the driving conditions, this approach tends to produce sub-optimal results. This paper presents a nonlinear MPC-based algorithm which incorporates the nonlinear kinematics of the Stewart platform within the MPC algorithm in order to increase the cueing fidelity and use maximum workspace. Furthermore, adaptive weights-based tuning is used to smooth the movement of the platform towards its physical limits. Full-track simulations were carried out and performance indicators were defined to objectively compare the response of the proposed algorithm with classical washout filter and linear MPC-based algorithms. The results indicate a better reference tracking with lower root mean square error and higher shape correlation for the proposed algorithm. Lastly, the effect of the adaptive weights-based tuning was also observed in the form of smoother actuator movements and better workspace use.

No abstract

The public acceptance of automated driving is influenced by multiple factors. Apart from safety being of top priority, comfort and time efficiency also have an impact on the popularity of automated vehicles. These two factors contradict each other as optimizing for one results in the degradation of the other. We investigate in this paper how such a multiobjective problem is approached by human drivers and by numerical optimization in the roundabout scenario, which is compact in size but complex to handle. The human drivers' behavior is first observed using naturalistic driving data. The average trajectories and distribution of peak accelerations were extracted after model-based fitting and removal of erroneous samples. The processed data is shared online as an openaccess dataset. Then, an optimization problem is formulated and solved to find the numerically optimal motion profile in terms of comfort and time efficiency. The weighted sum of travel time and discomfort is minimized. By adjusting the weight distribution, we present different motion profiles favoring optimal comfort, human-like acceleration magnitudes, and agility, respectively.

The benefits of automated driving can only be fully realized if the occupants are protected from motion sickness. Active suspensions hold the potential to raise the comfort level in automated passenger vehicles by enabling new functionalities in chassis control. One example is to actively lean the vehicle body toward the center of the corner to counteract the inertial lateral acceleration. Commonly known as curve tilting, the concept is deemed effective in reducing postural disturbance on the occupants and the visual-vestibular conflict when the occupants do not have an external view. We present in this article a nonlinear model predictive control (NMPC) method for the curve tilting functionality. The controller incorporates the nonlinear suspension forces in the prediction model to help achieve high tracking accuracy near the physical limit of the suspension system. The optimization process is accelerated with an explicit initialization method that is based on piecewise-affine (PWA) modeling and offline solution to an alternative optimal control problem (OCP). The controller is able to operate at 20 Hz in a hardware-in-the-loop (HIL) setup. Given sufficient computational resources, we observe a significant reduction in the lateral acceleration sensed by the passenger over a vehicle with passive suspensions, namely, by 46.5%, 25.4%, and 25.4% in the highway, rural, and urban driving scenarios, respectively. The NMPC also outperforms the baseline proportional-integralderivative (PID) controller by achieving lower tracking error, namely, by 12.9%, 16.4%, and 38.0% in the aforementioned scenarios.

Loss of lateral stability is still a major cause of road accidents in recent years. Nonlinear model predictive control (NMPC) is regarded as a powerful tool to improve vehicle safety by fully utilizing the tire-road friction. However, the computational load is excessive for on-board hardware, thus NMPC is yet distant from real-time implementation in vehicle control. To tackle the problem, this study proposes a method to improve the computational efficiency in NMPC. First, a lookup table of initial guess of the NMPC solution is calculated based on a piecewise-affine (PWA) approximation of tire's nonlinear behavior. Then, a local optimization process starting from the initial guess searches for the optimal control input, using perpendicular search plus line search method. The proposed method was examined through a set of numerical tests and in real-world scenarios using experimentally calibrated multibody model. The code was also tested on dSPACE for proof of real-time implementation. Results show a significant reduction in computational time, thus real-time implementation has been achieved with a huge margin. The loss in the accuracy of the optimal solution is negligible. The performance in improving vehicle safety is promising too, as the vehicle can be recovered from unstable motion with body slip angle up to 30 degrees.

Advanced passenger vehicles are complex dynamic systems that are equipped with several actuators, possibly including differential braking, active steering, and semi-active or active suspensions. The simultaneous use of several actuators for integrated vehicle motion control has been a topic of great interest in literature. To facilitate this, a technique known as control allocation (CA) has been employed. CA is a technique that enables the coordination of various actuators of a system. One of the main challenges in the study of CA has been the representation of actuator dynamics in the optimal CA problem (OCAP). Using model predictive control allocation (MPCA), this problem has been addressed. Furthermore, the actual dynamics of actuators may vary over the lifespan of the system due to factors such as wear, lack of maintenance, etc. Therefore, it is further required to compensate for any mismatches between the actual actuator parameters and those used in the OCAP. This is done by combining the MPCA solver with an online adaptive parameter estimation (APE) algorithm. In this study, an advanced solution to the OCAP is proposed by combining MPCA with APE. This solution coordinates differential braking and active front steering (AFS) of a passenger vehicle, to stabilize the lateral and yaw motion. The simulation results indicate that the APE+MPCA combination effectively accounts for actuator dynamics and actuator parameter mismatches. FIGURE 7 Yaw rate response-APE+MPCA.

Driving simulators are widely used for understanding human-machine interaction, driver behavior and in driver training. The effectiveness of simulators in these process depends largely on their ability to generate realistic motion cues. Though the conventional filter-based motion cueing strategies have provided reasonable results, these methods suffer from poor workspace management. To address this issue, linear MPC-based strategies have been applied in the past. However, since the kinematics of the motion platform itself is non-linear and the required motion varies with the driving conditions, this approach tends to produce sub-optimal results. This paper presents a nonlinear MPC-based algorithm which incorporates the nonlinear kinematics of the Stewart platform within the MPC algorithm in order to increase the cueing fidelity and utilize maximum workspace. Further, adaptive weights-based tuning is used to smoothen the movement of the platform towards its physical limits. Full-track simulations were carried out and performance indicators were defined to objectively compare the response of the proposed algorithm with classical washout filter and linear MPC-based algorithms. The results indicate a better reference tracking with lower root mean square error and higher shape correlation for the proposed algorithm. Lastly, the effect of the adaptive weights-based tuning was also observed in the form of smoother actuator movements and better workspace utilization.