UDMC: Unified Decision-Making and Control Framework for Urban Autonomous Driving with Motion Prediction of Traffic Participants

TL;DR

UDMC addresses the challenge of integrating decision-making and motion control for urban autonomous driving by formulating a receding-horizon optimal control problem that jointly optimizes trajectory tracking, control effort, and a comprehensive artificial potential field term encoding traffic rules and obstacle avoidance. The framework combines an interpretable APF-based description of traffic objects with an IGPR-based motion predictor to anticipate surrounding agents, feeding into a unified OCP solved via direct multiple shooting. Through CARLA simulations across ML-ACC, roundabout, signalized intersection, and mixed-T-junction scenarios, UDMC demonstrates improved safety (no collisions or rule violations) and efficiency relative to rule-based and learning-based baselines, with ablation and robustness analyses validating the contribution of motion prediction and PF design. The open-source implementation and demonstration on urban driving tasks suggest practical viability and provide a modular platform for extending unified decision-making and control in autonomous driving.

Abstract

Current autonomous driving systems often struggle to balance decision-making and motion control while ensuring safety and traffic rule compliance, especially in complex urban environments. Existing methods may fall short due to separate handling of these functionalities, leading to inefficiencies and safety compromises. To address these challenges, we introduce UDMC, an interpretable and unified Level 4 autonomous driving framework. UDMC integrates decision-making and motion control into a single optimal control problem (OCP), considering the dynamic interactions with surrounding vehicles, pedestrians, road lanes, and traffic signals. By employing innovative potential functions to model traffic participants and regulations, and incorporating a specialized motion prediction module, our framework enhances on-road safety and rule adherence. The integrated design allows for real-time execution of flexible maneuvers suited to diverse driving scenarios. High-fidelity simulations conducted in CARLA exemplify the framework's computational efficiency, robustness, and safety, resulting in superior driving performance when compared against various baseline models. Our open-source project is available at https://github.com/henryhcliu/udmc_carla.git.

Figures (7)

• Figure 1: An intelligent vehicle with the proposed driving framework is driving in a roundabout with congested traffic conditions. (a) The vehicle is merging into the roundabout when a surrounding vehicle is blocking another lane. (b) The vehicle is driving in the inner lane when there are two surrounding vehicles in front and one surrounding vehicle attempting to merge in.
• Figure 2: The proposed unified decision-making and control (UDMC) framework for urban autonomous driving. The arrows with different colors represent different kinds of information transmission routes.
• Figure 3: The designed potential functions for traversable and non-traversable lane markings, surrounding vehicles, and red traffic light signals. (a) $F_\text{CR}$ and $F_\text{NR}$ for crossable and non-crossable lanes. The potential field is summarized by two traversable and two non-traversable lane markings constituting a three-lane road structure. The lateral position of non-traversable markers and traversable markers are [5.25, -5.25] m and [1.75, -1.75] m, respectively. (b) $F_\text{TL}$ for traffic control signals. The PF of the red traffic light, where the parameters of the PF are set as $p_y^\text{ev} = 8.0$ m. (c) $F_\text{VC}$ for vehicles with circles wrapping. The PF is obtained by wrapping a vehicle using two circles, where the vehicle's position is set as $[p^k_x, p^k_y]=[3.0,0]$ m. (d) $F_\text{VE}$ for vehicles with an ellipsoid wrapping. The PF is obtained by wrapping a vehicle using an ellipsoid to construct the second version of vehicle PF, where $[p^k_x, p^k_y]=[4.0,0]$ m.
• Figure 4: Illustration of the effectiveness of the proposed VPF. (a) The VPF exists in an intersection, where the vehicle is crossing the intersection with the protection of the proposed VPF to avoid diverging too much from the target destination when surrounding vehicles are nearby. (b) The VPF exists before turning left, where the vehicle is waiting behind the leading vehicle, and driving to another road lane is prohibited by traffic rules.
• Figure 5: Simulation results by UDMC in four illustrative urban driving scenarios in CARLA. Four keyframes for each scenario are selected. The red dots above the surrounding vehicles are the predicted states of their future trajectory in the next 10 time steps generated by IGPR. In the Mix-T-U scenario, two pedestrians are crossing crosswalks on the way of the autonomous vehicle. Corresponding videos are accessible at https://www.youtube.com/watch?v=jftTsf1jXjU.

Theorems & Definitions (5)

• Remark 1
• Remark 2
• Remark 3
• Remark 4
• Remark 5