Don't Get Stuck: A Deadlock Recovery Approach

TL;DR

The paper tackles the problem of autonomous-vehicle deadlocks in constrained urban traffic by presenting a real-time recovery framework that blends Hybrid $A^ op$ path planning, Signal Temporal Logic ($STL$) constraints, and Model Predictive Path Integral ($MPPI$) control. It introduces an STL-based constraint layer within MPPI, adding penalties $oldsymbol{ ext{P}}_oldsymbol{ imes}$ and trajectory weights to steer toward safe, rule-compliant recovery trajectories under uncertainty, while maintaining a receding-horizon optimization. The approach is built on a pipeline where a modified Hybrid $A^ op$ planner generates an initial feasible path, which is then refined by STL-MPPI to satisfy spatiotemporal safety and traffic constraints through a cost function combining running costs, terminal costs, and STL penalties. Validation includes high-fidelity simulations and hardware experiments with 1/10-scale MuSHR vehicles, demonstrating improved safety, feasibility, and computational efficiency over a traditional MPC baseline, with practical implications for real-world autonomous driving in complex traffic scenarios.

Abstract

When multiple agents share space, interactions can lead to deadlocks, where no agent can advance towards its goal. This paper addresses this challenge with a deadlock recovery strategy. In particular, the proposed algorithm integrates hybrid-A$^\star$, STL, and MPPI frameworks. Specifically, hybrid-A$^\star$ generates a reference path, STL defines a goal (deadlock avoidance) and associated constraints (w.r.t. traffic rules), and MPPI refines the path and speed accordingly. This STL-MPPI framework ensures system compliance to specifications and dynamics while ensuring the safety of the resulting maneuvers, indicating a strong potential for application to complex traffic scenarios (and rules) in practice. Validation studies are conducted in simulations and on scaled cars, respectively, to demonstrate the effectiveness of the proposed algorithm.

Figures (11)

• Figure 1: In the depicted scenario, the ego vehicle, marked in red, faces a deadlock within a construction zone while following its predefined global path, indicated by a yellow line. To resolve the deadlock, the vehicle needs to create a new path that circumvents the construction area, avoids collisions with other vehicles, and allows it to continue on its intended route.
• Figure 2: In this scenario, the ego vehicle confronts a deadlock caused by a motorcycle crash. To escape, it must perform complex backward and forward maneuvers to safely navigate around the crash site and continue its journey without colliding with other vehicles.
• Figure 3: The proposed system integrates traffic, lane, and spatial data into a Hybrid A$^\star$ planner to formulate a path for an autonomous vehicle. This path informs the MPPI controller, which integrates traffic rules and vehicle constraints as STL specifications. The controller then outputs commands that align with these rules while maintaining proximity to the planned path.
• Figure 4: The navigable space in the system is discretized into sparse grids along the lanes and finer grids around obstacles. This approach ensures that even in tighter spaces, a feasible path is always available.
• Figure 5: The Hybrid A$^\star$ algorithm generates a feasible path for vehicles by adhering to vehicle dynamics and spatial constraints, such as staying within lane boundaries and maintaining safe distances from obstacles.

Theorems & Definitions (3)

• Remark 1
• Remark 2
• Remark 3