Search

This paper considers computationally efficient planning of reference trajectories for autonomous on-road vehicles in a cooperative setting. The basic element of the approach is the notion of so-called maneuvers, which allow to cast the nonlinear and non-convex planning task into a highly structured optimal control problem. This can be solved quite efficiently, but not fast enough for online operation when considering nonlinear vehicle models. Therefore, the approach proposed in this paper aims at approximating solutions using a supervised learning approach: First, training data are generated by solving optimal control problems and are then used to train a neural network. As is demonstrated for a cooperative overtaking maneuver, this approach shows good performance, while (contrasting approaches like reinforcement learning) requiring only low training effort.

In the context of trajectory planning for autonomous vehicles, a widely used vehicle model relies on linear integrator dynamics. We consider planning with this model type, with a focus on the requirement to account for curved road topologies. As our analysis reveals, this generally gives rise to non-convex, coupled constraints on the vehicle’s states and inputs, which impedes computationally efficient planning. We propose a method to resolve this issue by modification of the non-convex constraints. This modification is based on inner approximations of sub-level sets of nonlinear functions, which are obtained by quantifier elimination. The efficacy of the method is demonstrated in two example scenarios.

This paper proposes a framework for generation of collision-free reference trajectories in a cooperative multi-agent setting. The approach is hierarchical: a high-level controller schedules groups of cooperating agents, for which trajectories are then determined by a lower-level trajectory planner. Admissible behaviors of a cooperative group are encoded by so-called maneuvers, which are modeled by hybrid automata. This allows to plan trajectories by solving hybrid optimal control problems. Controllable sets characterize the solution sets of these problems and are used for quick assessment of feasibility prior to planning. This enables the high-level controller to quickly assess the feasibility of different maneuver options and cooperative groups. Special emphasis is put on safety and feasibility of the framework. The efficacy of the approach is demonstrated by simulation of a highway entry scenario for autonomous vehicles.

This paper addresses the design of sparse, H2-optimized dynamic output feedback controllers for distributed systems. These systems consist of several physically remote subsystems, which are assumed to be controlled via local controllers. Depending on the topology of a communication network, subsets of these controllers may exchange information. In this setting, a trade-off between control performance and communication costs exists: while a network which connects more controllers increases control performance, establishing communication links also causes higher costs. In the past, the Alternating Direction Method of Multipliers (ADMM) has been applied to the design of state feedback controllers that optimize this trade-off. Due to the practical relevance of output feedback, this paper provides extensions necessary for application of the ADMM to the output feedback case. Motivated by deeper insights into the non-convex, non-smooth nature of the problem, the HANSO optimization package is applied to the controller performance optimization subproblem of the ADMM. The devised procedure is successfully applied to two example systems.

Motion planning of automated vehicles requires dynamical models to ensure that obtained trajectories are drivable. An often overlooked aspect is that motion planning is usually done using simplified models, which do not always sufficiently conform to the real behavior of vehicles. Thus, collision avoidance and drivability is not necessarily ensured. We address this problem by modeling vehicles as differential inclusions composed of simple dynamics and set-based uncertainty; conformance testing is used to determine the required uncertainty. To quickly provide the set of solutions for these uncertain models, we use pre-computed reachable sets (i.e., the union of all possible solutions) for pre-selected motion primitives. The reachable sets of vehicles are obtained through the novel combination of optimization techniques and reachability analysis in the controller synthesis — they enable us to guarantee safety by checking their mutual non-intersection for consecutive time intervals. The benefits of our approach are demonstrated by numerical experiments.

This paper aims at cooperatively resolving conflicts arising between several fully-autonomous, communicating vehicles in general on-road traffic scenarios. A hierarchical control scheme with two main units is proposed: A short-term planning unit determines obstacle-avoiding reference trajectories in a receding-horizon-fashion, based on solving mixed-integer quadratic programs (MIQP) for simplified vehicle dynamics. A novel cooperative conflict resolution scheme is proposed which orchestrates the plans of neighboring vehicles, trying to minimize a common cost function. Tracking controllers based on model predictive control (MPC) are then employed to track the references using a nonlinear vehicle model. The functionality of the proposed method is demonstrated in a simulation example.