Search

Highly automated driving requires the use of multiple sensors for reliable tracking functionality. In response to the requirement, the proposed method modifies the conventional Interacting Multiple Model (IMM) filter to fuse multi-sensor data by utilizing the independence of observations. In addition, the proposed IMM is integrated with a Centralized Kalman Filter (CKF) that ensures track continuity against sensor failures, providing optimal state estimates. When tracking objects in a moving reference frame, such as in autonomous vehicles, onboard sensor measurements represent relative values, making it challenging to estimate the actual motion of objects. While transforming states to a global coordinates is a solution, the solution can arise another problem where the tracking results depends on the status and performance of the localization. Therefore, to tackle the problem, a track compensation algorithm utilizing a hybrid coordinate system is proposed. The actual motions of objects are estimated based on errors between the track state and the associated measurement. The performance of the proposed algorithms is demonstrated using experimental scenario data conducted with an actual vehicle.

Precise localization is essential for reliable autonomous driving. Traditionally, many systems have turned to lane level map matching techniques utilizing High Definition Maps (HD-Maps). However, it is necessary to have considerate efforts tied to the continuous availability and timeliness of these HD-Maps. Taking this into account, this paper explores an alternative approach, focusing on landmark-based odometry and a filtering-based technique. The proposed localization algorithm is integrated with the GPS and the odometry in a loosely coupled approach. Because it is important to obtain an initial pose in order to utilize the odometry, the initial odometry rotation is estimated from the filtering method, as an alternative to optimization methods, to improve the global consistency in localization. In addition, because specific landmark-based odometry, like lidar odometry, can become vulnerable in geometrically repetitive scenarios such as in tunnels, this study considers the change in vehicle speed and updates the landmark-based odometry, aiming to address the limitations. Through experimental tests in normal scenarios and challenging scenarios like tunnels, it is demonstrated that our method offers certain benefits over the existing techniques.

This paper proposes a behavior planning algorithm for self-driving vehicles to handle lane keeping, speed control considering inter-vehicle space, and collision avoidance under uncertainty. The behavior planning approach is structured as a hierarchically organized Markov Decision Process (MDP) comprising two components: the path planning MDP and the speed profiling MDP. The path planning MDP generates multiple path candidates using lane-change path data collected from human drivers. Evaluation of each path candidate is based on reward and penalty terms. The path planning MDP spans and updates considering the predicted simulation time of the path candidate. Subsequently, the speed profiling MDP determines the optimal sequence of speeds for the host vehicle on the planned path. Evaluation of the path and speed profile utility is performed using reward and penalty terms based on the current states of vehicles and road structure. A unique aspect of this approach is the incorporation of uncertainty-aware collision risk and interaction-aware gap penalty, which account for the uncertainty of perception and traffic motion. Various cut-in scenarios are presented in simulations to demonstrate the effectiveness of the proposed algorithm.

The tracking accuracy of nearby vehicles determines the safety and feasibility of driver assistance systems or autonomous vehicles. Recent research has been active to employ additional sensors or to combine heterogeneous sensors for more accurate tracking performance. Especially, autonomous driving technologies require a sensor fusion technique that considers various driving environments. In this research, a novel method for high-level data fusion is proposed to improve the accuracy of tracking surrounding vehicles. In response to the changing driving environment, the locations of the vehicles are estimated in real-time using an adaptive track-to-track fusion technique and an interacting multiple model filter. Asynchronous measurements from multiple sensors such as radar, camera, and LiDAR, are utilized for the estimation. For each sensor, two motion models representing the vehicle’s movement are applied to increase the estimation accuracy. Utilizing a multimodal network-based track-to-track fusion approach, it combines the estimates of the target vehicle position from each sensor into a single estimate. The inputs of the network are intended to determine the reliability of each sensor, considering the driving conditions that may affect sensor accuracy. Also, multiple embeddings in the network are created so that the corresponding data maintains its relevance and enables the real-time computing. The proposed method is verified using real driving data collected from various environments.

A robust and high-performance lateral control system is necessary for autonomous bus driving in complex downtown roads. In this paper, a Sliding Mode (SM) controller with an uncertainty observer is proposed to reduce the chattering phenomenon while ensuring robustness and maintaining the control performance. A stiffness gain is introduced in the cornering stiffness of rear wheels to compensate for the model uncertainty and external disturbance. For lane keeping and changing maneuvers, an optimal path planning algorithm is designed by considering dynamic constraints of the bus. The lane departure of bus is prevented with minimizing the lateral jerk through the sampling-based path regeneration process. The proposed lateral control system is compared to other conventional SM based controllers in the TRUCKSIM simulator. In an experiment with a real bus, lane keeping maneuver is conducted and compared with experienced human driver in the same driving course. Lane change maneuver is also tested with real bus to perform smooth lane transition while satisfying the lateral acceleration and yaw rate limitations.

This paper presents an efficient hierarchical motion planning framework with a long planning horizon for autonomous driving in structured environments. A 3D motion planning with time information is a non-convex problem because there exists more than one local minimum point and various constraints such as roads and obstacles. Accordingly, to deal with high computational complexity and the problem of falling into a local minimum in an enormous solution space, a decoupled method is utilized, that consists of two steps: Long-term planning and short-term planning. First, the long-term planner provides reasonable far-sighted behavior through two processes. In the first process, a rough path that includes a driving strategy is generated in the 2D spatial space. Then, the jump point search algorithm is utilized with time information on the path to reduce the computational burden of A*, giving an acceptable quality of solution at the same time. In this step, a safe, comfortable, and dynamically feasible trajectory is generated. Next, the short-term planner optimizes a short-sighted trajectory using particle swarm optimization. In this method, a steering angle set is encoded as a particle, resulting in a safe, comfortable, and kinodynamically feasible trajectory. The proposed algorithm is implemented and evaluated in a range of vehicle-in-the-loop simulation scenarios, which include various virtual static and dynamic obstacles generated by Intelligent Driver Model. In the evaluation results, the proposed method reduced the computation time by up to 0.696 s with increasing the step cost by up to about 3%. The proposed algorithm is executed every 100 ms for a planning horizon of 10 seconds, and the average computation time is 31 ms with the worst-case computation time of 94 ms.

An integration scheme for sideslip angle estimation is proposed where a deep neural network and a simple kinematics-based model are combined in an unscented Kalman filter. The deep neural network contains two modules: a sensor filtering network designed to overcome the limitations of the kinematics-based model and a deep ensemble network to estimate the sideslip angle and its uncertainty. Both networks use recurrent neural networks with long short-term memory to analyze sequential sensor data. The networks were trained using only input signal sets that can be obtained from on- board sensor measurements. The filtering network reduces the noise and bias of the input signals to match the model used for the unscented Kalman filter. Next, the initial estimate and its uncertainty obtained from the deep ensemble network are utilized as a new measurement in the unscented Kalman filter, inducing an adaptive measurement variance. The algorithm was verified through both simulation and experiment, and the results demonstrate the effectiveness of the proposed algorithm.

Trajectory prediction of the ego vehicle is essential for advanced driver assistance systems to function properly. By recognizing various driving styles and predicting trajectories reflecting them, the prediction performance is enhanced, and a personalized trajectory can be generated. Therefore, we propose to combine driving style recognition and trajectory prediction tasks using only in-vehicle CAN-bus sensor data for possible application to normal vehicles. The DeepConvLstm network was utilized for driving style recognition, and a generative-based model was used for trajectory prediction. The classified driving style was added as a condition to the network. In addition, the past trajectory of the ego vehicle is estimated and utilized as an additional input for performance improvement. The performance of the proposed method is analyzed in terms of the root mean squared error (RMSE) and mean absolute error (MAE) compared with the case wherein the driving style and the past trajectory are not conditioned or given, respectively. The results demonstrate the effectiveness of the proposed method.

Prediction of surrounding vehicles accurately is an essential prerequisite for safe autonomous driving. Trajectory prediction methods can be classified into physics-, maneuver-, or learning-based methods. Learning-based methods have been studied extensively in recent years because it effectively exploits the road information and interactions among vehicles. However, learning-based methods perform poorly in unseen environments that were not considered during training and provide unreasonable results such as inconsistent trajectories according to road geometry. In this paper, to address this problem, a hybrid model that combines a learning-based model with physics- and maneuver-based models according to their uncertainties is proposed. The deep ensemble technique is also used to estimate the uncertainty of the learning-based method. Because the deep ensemble tends to show a large variance in unseen environments, this method is used to determine whether to use a hybrid model. The proposed method is trained and validated using the Lyft l5 dataset, the real environment vehicle driving data containing several types of intersections.

In this article, a unified lane-keeping and forward collision avoidance system is proposed for semi-trailer trucks. Kinematic and dynamic models are described for lateral motion of the semi-trailer truck. The variation of the cornering stiffness is considered because its value depends on driving conditions such as vehicle speed and vertical load. The proposed system consists of the hitch angle estimation algorithm of the semi-trailer truck and the Model Predictive Controller (MPC) for the lateral motion of the semi-trailer truck. In the hitch angle estimation, the disturbance observer is utilized to compensate the lateral uncertainties and the estimation performance is validated in simulations and experiments. In the controller design, an MPC controller is designed both for lane keeping and collision avoidance, but their constraints in lateral motion are considered separately. In constructing the objective functions, constraint smoothing is applied to improve the feasibility of the semi-trailer truck motion. A simplification method is also introduced to reduce the computational complexity of the MPC and to improve the real-time performance. The proposed system is evaluated in simulations for lane keeping and collision avoidance, respectively.

A planar motion control system is proposed for autonomous vehicles not only to follow the lanes, but also to avoid collisions by braking, accelerating, and steering. The supervisor is designed first to determine the desired speed and the risk of the maneuvering due to road boundaries and obstacles. In order to allow lane changes on multi-lane roads, the model predictive controller is formulated based on the probabilistic non-convex optimization. The micro-genetic algorithm is applied to calculate the target speed and target steering angle in real time. A software-in-the-loop unit is constructed with the Rapid Control Prototyping device in the vehicle communication environment. The performance of the proposed system is verified for various collision avoidance scenarios and the simulation results demonstrate the safe and effective driving performance of autonomous vehicles with no collision on multi-lane road.

Deep learning is a fast-growing field of research, in particular, for autonomous application. In this study, a deep learning network based on various sensor data is proposed for identifying the roads where the vehicle is driving. Long-Short Term Memory (LSTM) unit and ensemble learning are utilized for network design and a feature selection technique is applied such that unnecessary sensor data could be excluded without a loss of performance. Real vehicle experiments were carried out for the learning and verification of the proposed deep learning structure. The classification performance was verified through four different test roads. The proposed network shows the classification accuracy of 94.6% in the test data.