Safety is one of the key requirements for automated vehicles and fault diagnosis is an effective technique to enhance the vehicle safety. The model-based fault diagnosis method models the fault into the system model and estimates the faults by observer. In this article, to avoid the complexity of designing observer, we investigate the problem of steering actuator fault diagnosis for automated vehicles based on the approach of model-based support vector machine (SVM) classification. The system model is utilized to generate the residual signal as the training data and the data-based algorithm of the SVM classification is employed to diagnose the fault. Due to the phenomena of data unbalance induced poor performance of the data-driven method, an undersampling procedure with the approach of linear discriminant analysis and a threshold adjustment using the algorithm of grey wolf optimizer are proposed to modify and improve the performance of classification and fault diagnosis. Various comparisons are carried out based on widely used datasets. The comparison results show that the proposed algorithm has superiority on the classification over existing methods. Experimental results and comparisons of an automated vehicle illustrate the effectiveness of the proposed algorithm on the steering actuator fault diagnosis.

Fault Diagnosis of an Autonomous Vehicle With an Improved SVM Algorithm Subject to Unbalanced Datasets

This paper describes lane change motion planning with a combination of probabilistic and deterministic prediction for automated driving under complex driving circumstances. The autonomous lane change should arrive safely at the destination. The subject vehicle needs to perceive and predict the behaviors of other vehicles with sensors. From the information of other vehicles, a collision probability is defined using a reachable set of uncertainty propagation. In addition, the lane change risk is monitored using predicted time-to-collision and safety distance to guarantee safety in lane change behavior. A safe driving envelope is defined as constraints based on the combinatorial prediction (probabilistic and deterministic) of the behavior of surrounding vehicles. To obtain the desired steering angle and longitudinal acceleration to maintain the automated driving vehicle under constraints, a stochastic model-predictive control problem is formulated. The proposed model has been evaluated by performing lane change simulations in MATLAB/Simulink, while considering the effect of combination prediction. Also, the proposed algorithm has been implemented on a test vehicle. The simulation and test results show that the proposed algorithm can handle complicated lane change scenarios, while guaranteeing safety.

Stochastic Model-Predictive Control for Lane Change Decision of Automated Driving Vehicles

The safety and resilience of fully autonomous vehicles (AVs) are of significant concern, as exemplified by several headline-making accidents. While AV development today involves verification, validation, and testing, end-to-end assessment of AV systems under accidental faults in realistic driving scenarios has been largely unexplored. This paper presents DriveFI, a machine learning-based fault injection engine, which can mine situations and faults that maximally impact AV safety, as demonstrated on two industry-grade AV technology stacks (from NVIDIA and Baidu). For example, DriveFI found 561 safety-critical faults in less than 4 hours. In comparison, random injection experiments executed over several weeks could not find any safety-critical faults.

ML-Based Fault Injection for Autonomous Vehicles: A Case for Bayesian Fault Injection

This paper proposes AV-FUZZER, a testing framework, to find the safety violations of an autonomous vehicle (AV) in the presence of an evolving traffic environment. We perturb the driving maneuvers of traffic participants to create situations in which an AV can run into safety violations. To optimally search for the perturbations to be introduced, we leverage domain knowledge of vehicle dynamics and genetic algorithm to minimize the safety potential of an AV over its projected trajectory. The values of the perturbation determined by this process provide parameters that define participants’ trajectories. To improve the efficiency of the search, we design a local fuzzer that increases the exploitation of local optima in the areas where highly likely safety-hazardous situations are observed. By repeating the optimization with significantly different starting points in the search space, AV-FUZZER determines several diverse AV safety violations. We demonstrate AV-FUZZER on an industrial-grade AV platform, Baidu Apollo, and find five distinct types of safety violations in a short period of time. In comparison, other existing techniques can find at most two. We analyze the safety violations found in Apollo and discuss their overarching causes.

AV-FUZZER: Finding Safety Violations in Autonomous Driving Systems

This paper provides a new solution for path following control of autonomous ground vehicles. $\mathcal {H}_{2}$ control problem is considered to attenuate the effect of the road curvature disturbance. To this end, we formulate a standard model from the road-vehicle dynamics, the a priori knowledge on the road curvature, and the path following specifications. This standard model is then represented in a Takagi–Sugeno fuzzy form to deal with the time-varying nature of the vehicle speed. Based on a static output feedback scheme, the proposed method allows avoiding expensive vehicle sensors while keeping the simplest control structure for real-time implementation. The concept of $\mathcal {D}-$ stability is exploited using Lyapunov stability arguments to improve the transient behaviors of the closed-loop vehicle system. In particular, the physical upper and lower bounds of the vehicle acceleration are explicitly considered in the design procedure via a parameter-dependent Lyapunov function to reduce drastically the design conservatism. The proposed $\mathcal {H}_{2}$ design conditions are expressed in terms of linear matrix inequalities (LMIs) with a single line search parameter. The effectiveness of the new path following control method is clearly demonstrated with both theoretical illustrations and hardware experiments under real-world driving situations.

Fuzzy Static Output Feedback Control for Path Following of Autonomous Vehicles With Transient Performance Improvements

This paper presents a novel approach for vision-based Simultaneous Localization and Mapping (SLAM) on a high curved track without pose-based landmarks. The proposed approach combines an Iterative Closest Points (ICP) method and Stochastic Gradient Descent (SGD) optimization and comprises four main steps. First, a Kalman filter with a simple circular lane model is used to estimate the road curvature using images from a front camera. Then, the vehicle position and orientation are reconstructed by using the yaw rate and longitudinal speed from inertial sensors. Drift and misalignment of the constructed map are corrected using ICP under the assumption that the vehicle continuously travels the same track. The final map is obtained using SGD optimization, which enforces curvature matching. We evaluate the performance of the proposed algorithm with an environment of the winding track of Hyundai-Kia California Proving Ground (CPG) located in Southern California and the customized ellipsoidal track. The experimental results show the effectiveness of the proposed approach.

Vision-Based Race Track SLAM Based Only on Lane Curvature

This paper represents an automated driving control algorithm in urban traffic situation. In order to achieve a development of a highly automated driving control algorithm in urban environments, the research issues can be classified into two things. One of the issues is to determine a safe driving envelope with the consideration of probable risks and the other is to achieve robustness of control performance under disturbances and model uncertainties. While human drivers maneuver a vehicle, they determine appropriate steering angle and acceleration based on the predictable trajectories of the surrounding vehicles. Therefore, not only current states of surrounding vehicles but also predictable behaviors of surrounding vehicles and potential obstacles should be considered in designing an automated driving control algorithm. In order to analyze the probabilistic behaviors of surrounding vehicles, we collected driving data on a real road. Then, in order to guarantee safety to the possible change of traffic situation surrounding the subject vehicle during a finite time-horizon, the safe driving envelope which describes the safe driving condition over a finite time horizon is defined in consideration of probabilistic prediction of future positions of surrounding vehicles and potential obstacles. Since an automated driving control algorithm is required to operate in a wide operating region and limit the set of permissible states and inputs, a model predictive control (MPC) approach has been used widely in designing an automated driving control algorithm. MPC approach uses a dynamic model of the vehicle to predict the future states of the system and determines optimal control sequences at each time step to minimize a performance index while satisfying constraints based on the predicted future states. Since the solving nonlinear optimization problem has computational burden, we design an architecture which decides a desired steering angle and longitudinal acceleration parallel to reduce the computational load. For the guarantee of the robustness of control performance, a robust invariant set is used to ensure robust satisfaction of vehicle states and constraints against disturbances and model uncertainties. The effectiveness of the proposed control algorithm is evaluated by comparing between human driver data and proposed algorithm.

Jongsang Suh

Papers

Vision-Based Race Track SLAM Based Only on Lane Curvature

Probabilistic prediction based automated driving control in urban traffic situation

Lane change for self-driving in highly dense traffic using motion based uncertainty propagation