AbstractThere is currently a strongly growing interest in obtaining optimal control solutions for vehicle manoeuvres, both order to understand behaviour and, perhaps more importantly, devise improved safety systems, either by direct deployment of the or including mimicked driving techniques professional drivers. However, it non-trivial find right combination models, optimisation criteria, and tools get useful results above purposes. Here, platform investigation these aspects developed based...

We present a novel approach to learning online the tire stiffness and vehicle state using only wheel-speed inertial sensors. The deviations from nominal values are treated as Gaussian disturbance acting on vehicle. formulate Bayesian approach, in which we leverage particle filtering marginalization concept estimate computationally efficient way tire-stiffness parameters state. In estimation model, process measurement noises dependent each other, an account for dependence. Our algorithm...

We propose an adaptive nonlinear model predictive control (NMPC) for vehicle tracking control. The controller learns in real time a tyre force to adapt varying road surface that is only indirectly observed from the effects of forces determining dynamics. Learning entire data would require driving unstable region dynamics with prediction has not yet converged. Instead, our approach combines NMPC noise-adaptive particle filter state and stiffness estimation pre-determined library models....

We consider the object recognition problem in autonomous driving using automotive radar sensors. Comparing to Lidar sensors, is cost-effective and robust all- weather conditions for perception driving. However, signals suffer from low angular resolution precision recognizing surrounding objects. To enhance capacity of radar, this work, we exploit temporal information successive ego-centric bird-eye-view image frames recognition. leverage consistency an object's existence attributes (size,...

Motion planning under differential constraints is one of the canonical problems in robotics. State-of-the-art methods evolve around kinodynamic variants popular sampling-based algorithms, such as Rapidly-exploring Random Trees (RRTs). However, there are still challenges remaining, for example, how to include complex dynamics while guaranteeing optimality. If open-loop unstable, exploration by random sampling control space becomes inefficient. We describe CL-RRT#, which leverages ideas from...

This paper discusses some of the current state-of-the-art and remaining challenges in research on path planning vehicle control autonomous vehicles. Reliable is fundamental for proper operation an vehicle. Typically, planner relies incomplete model surroundings to generate a reference trajectory, used as input controller that tracks this trajectory. Depending how much complexity put into path-planning block, vehicle-control blocks can be viewed independent each other, connected or merged one...

Joint wheel-slip and vehicle-motion estimation is considered, based on measurements from wheel encoders, an inertial measurement unit, a global positioning system (GPS). The proposed strategy effectively employs the Rao-Blackwellized particle-filtering framework using kinematic model. Key variables in active safety systems, such as longitudinal velocity, roll angle, slip for all four wheels, are estimated. results demanding field test show efficacy of approach; velocity can be estimated with...

We present a gradient-based meta-learning framework for rapid adaptation of neural state-space models (NSSMs) black-box system identification. When applicable, we also incorporate domain-specific physical constraints to improve the accuracy NSSM. The major benefit our approach is that instead relying solely on data from single target system, utilizes diverse set source systems, enabling learning limited data, as well with few online training iterations. Through benchmark examples,...

For control architectures of autonomous and semi-autonomous driving features, we design a vehicle steering controller with limited preview ensuring that the constraints are satisfied, any piecewise clothoidal trajectory, is possibly generated by path planner or supervisory algorithm satisfies on desired yaw rate change rate, tracked within preassigned lateral error bound. The based computing non-maximal, yet polyhedral, robust invariant (RCI) set for system subject to bounded disturbances...

Steering control for autonomous vehicles on slippery road conditions, such as snow or ice, results in a highly nonlinear and therefore challenging online problem, which model predictive (NMPC) schemes have shown to be promising approach. NMPC allows deal with the vehicle dynamics well system limitations geometric constraints rather natural way, given desired trajectory that can provided by supervisory algorithm path planning. Our aim is study real-time feasibility of NMPC-based steering an...

This paper describes a probabilistic method for realtime decision making and motion planning autonomous vehicles. Our approach relies on the fact that driving road networks implies apriori defined requirements planner should satisfy. Starting from an initial state of vehicle, map, obstacles in region interest, goal region, we formulate motion-planning problem as nonlinear non-Gaussian estimation problem, which solve using particle filtering. We assign probabilities to generated trajectories...

The invariant-set motion planner uses a collection of safe sets to find collision-free path through an obstacle-filled environment [1]-[4]. This article extends the systems with persistently varying disturbances and parametric model uncertainty. is accomplished by replacing previously used positive invariant robust sets. Since uncertainty obfuscates relationship between in state space, references obstacles output we reformulate dynamics velocity form so that system appears directly modified...

This paper presents a data-driven measurement model for extended object tracking (EOT) with automotive radar. Specifically, the spatial distribution of radar measurements is modeled as hierarchical truncated Gaussian (HTG) structural geometry parameters that can be learned from training data. The HTG provides an adequate resemblance to real-world measurements. Moreover, large-scale datasets leveraged learn geometry-related and offload computationally demanding parameter estimation state...

This paper develops a method for safe lane changes. We leverage feedback control and constraint-admissible positively invariant sets to guarantee collision-free closed-loop trajectory tracking. Starting from an initial state of the vehicle obstacles in region interest, our steers desired while satisfying constraints associated with future motion respect vehicle. connect using equilibrium points dynamics, where are used transitions between points. An autonomous highway-driving example...

The system design of an autonomous vehicle encompasses numerous different interconnected sensing and control algorithms that can be devised in several ways, the has to extensively tested verified before employed on roads. Full-scale testing such a is complex due involved time effort, cost aspects, safety considerations. In this tutorial paper, we give overview design, implementation, stack vehicles, based our research motion planning control. We use scaled vehicles as part verification...

We present an algorithm for steering the output of a linear system from feasible initial condition to desired target position, while satisfying input constraints and nonconvex constraints. The is generated by collection local state-feedback controllers. path-planning selects appropriate controller using graph search, where nodes are controllers edges indicate when it possible transition one another without violating or two methods computing first uses fixed-gain scales its positive invariant...

This article describes a method for real-time integrated motion planning and control aimed at autonomous vehicles. Our leverages feedback control, positive invariant sets, equilibrium trajectories of the closed-loop system to produce track that are collision-free with guarantees according vehicle model. jointly steers target region controls velocity while satisfying constraints associated future surrounding obstacles. We develop receding-horizon implementation policy verify in both simulated...

One of the key technologies to safely operate self-driving vehicles is threat assessment other in neighborhood a vehicle. Threat algorithms must be capable predicting future movement vehicles. Many algorithms, however, predict trajectories based only on model dynamics and environment, which implies that they sometimes make too conservative predictions. This work reduces this conservativeness by capturing driver intention using random-forests classifier. Then, algorithm computes possible with...

This paper introduces the planning algorithm Sa-feRRT, which extends rapidly-exploring random tree (RRT) by using feedback control and positively invariant sets to guarantee collision-free closed-loop path tracking. The SafeRRT steers output of a system from feasible initial value desired goal, while satisfying input constraints non-convex constraints. constructs local state-feedback controllers, each with randomly sampled reference equilibrium corresponding set. indicate when it is possible...