End-to-End Intersection Handling using Multi-Agent Deep Reinforcement Learning

Capasso, Alessandro Paolo; Maramotti, Paolo; Dell'Eva, Anthony; Broggi, Alberto

doi:10.48550/arxiv.2104.13617

Cited by 3 publications

(5 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead of making simple assumptions, another way to consider traffic rules is to explicitly incorporate information into the state space or reward function of the DRL agent. In [13], stop lines and yield lines are represented in the state space using a grid map with different colors. The positions of other vehicles and their priority levels are also embedded within the state space.…”

Section: Ego Vehiclementioning

confidence: 99%

Safe and Rule-Aware Deep Reinforcement Learning for Autonomous Driving at Intersections

Zhang

Kacem

Hinz

et al. 2022

2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC)

View full text Add to dashboard Cite

Driving through complex urban environments is a challenging task for autonomous vehicles (AVs), as they must safely reach their mission goal, and react properly to traffic participants while obeying traffic rules. Deep reinforcement learning (DRL) is a promising method to generate driving policies for AVs because it can explore complex environments and learn suitable reactions. In this work, we present a DRL algorithm for AVs to handle intersection scenarios while considering traffic rules. Furthermore, we enhance the safety of our DRL algorithm's decisions by introducing a safety checker based on a responsibility-sensitive safety (RSS) model. Evaluations show that our DRL algorithm outperforms the baseline method by driving safely to reach the mission goal while obeying the traffic rules at an intersection.

show abstract

Section: Ego Vehiclementioning

confidence: 99%

Safe and Rule-Aware Deep Reinforcement Learning for Autonomous Driving at Intersections

Zhang

Kacem

Hinz

et al. 2022

2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC)

View full text Add to dashboard Cite

show abstract

“…4) Deep learning: Deep learning methods involving reinforcement learning [18] and recurrent neural networks [19] learn a planning and control policy, to be used by agents approaching an intersection. In practice, these policies do not generalize well to different environments and often do not provide guarantees in terms of safety and fairness.…”

Section: A Related Workmentioning

confidence: 99%

GAMEOPT: Optimal Real-time Multi-Agent Planning and Control for Dynamic Intersections

Suriyarachchi¹,

Chandra²,

Baras³

et al. 2022

Preprint

View full text Add to dashboard Cite

We propose GAMEOPT: a novel hybrid approach to cooperative intersection control for dynamic, multi-lane, unsignalized intersections. Safely navigating these complex and accident prone intersections requires simultaneous trajectory planning and negotiation among drivers. GAMEOPT is a hybrid formulation that first uses an auction mechanism to generate a priority entrance sequence for every agent, followed by an optimization-based trajectory planner that computes velocity controls that satisfy the priority sequence. This coupling operates at real-time speeds of less than 10 milliseconds in high density traffic of more than 10, 000 vehicles/hr, 100× faster than other fully optimization-based methods, while providing guarantees in terms of fairness, safety, and efficiency. Tested on the SUMO simulator, our algorithm improves throughput by at least 25%, time taken to reach the goal by 75%, and fuel consumption by 33% compared to auction-based approaches and signaled approaches using traffic-lights and stop signs.

show abstract

“…We evaluate our method by comparing it with state-ofthe-art planning methods [30], [23], [38], [25], [19], [2] for unsignaled and uncontrolled environments and show a maximum reduction in the number of collisions and deadlocks by up to 30%. Additionally, we compare our algorithm with an ablated version that does not use turn-based orderings and show that the time taken for all agents to navigate the scenarios increases in the latter case.…”

Section: A Main Contributionsmentioning

confidence: 99%

“…In Table I, we compare our approach with the current stateof-the-art in navigating unsignaled intersections, roundabouts, and merging scenarios on the basis of optimality guarantees, multi-agent versus single-agent planning (MAP), description of action space (AS), incentive compatibility (IC), and realworld applicability. DRL-based methods [2], [19], [20], [25], [26] learn a navigation policy using the notion of expected reward received by an agent from taking a particular action in a particular state. This policy is learned from trajectories obtained via traffic simulators using Q-learning [27] and is very hard as well as expensive to train.…”

Section: Prior Workmentioning

confidence: 99%

“…Furthermore, these methods discussed so far are intended for single-agent navigation. Capasso et al [2] use additional signals such as traffic signs (stop, yield, none) to regulate the movement and actions of other agents. In terms of real world applications, Kai et al [20] learn a unified policy for multiple tasks and also demonstrate their approach on a real robot.…”

Section: Prior Workmentioning

confidence: 99%

See 1 more Smart Citation

GamePlan: Game-Theoretic Multi-Agent Planning with Human Drivers at Intersections, Roundabouts, and Merging

Chandra¹,

Manocha²

2021

Preprint

View full text Add to dashboard Cite

We present a new method for multi-agent planning involving human drivers and autonomous vehicles (AVs) in unsignaled intersections, roundabouts, and during merging. In multi-agent planning, the main challenge is to predict the actions of other agents, especially human drivers, as their intentions are hidden from other agents. Our algorithm uses game theory to develop a new auction, called GAMEPLAN, that directly determines the optimal action for each agent based on their driving style (which is observable via commonly available sensors like lidars and cameras). GAMEPLAN assigns a higher priority to more aggressive or impatient drivers and a lower priority to more conservative or patient drivers; we theoretically prove that such an approach, although counter-intuitive, is gametheoretically optimal. Our approach successfully prevents collisions and deadlocks. We compare our approach with prior state-of-the-art auction techniques including economic auctions, time-based auctions (first-in first-out), and random bidding and show that each of these methods result in collisions among agents when taking into account driver behavior. We additionally compare with methods based on deep reinforcement learning, deep learning, and game theory and present our benefits over these approaches. Finally, we show that our approach can be implemented in the real-world with human drivers.

show abstract

End-to-End Intersection Handling using Multi-Agent Deep Reinforcement Learning

Cited by 3 publications

References 0 publications

Safe and Rule-Aware Deep Reinforcement Learning for Autonomous Driving at Intersections

Safe and Rule-Aware Deep Reinforcement Learning for Autonomous Driving at Intersections

GAMEOPT: Optimal Real-time Multi-Agent Planning and Control for Dynamic Intersections

GamePlan: Game-Theoretic Multi-Agent Planning with Human Drivers at Intersections, Roundabouts, and Merging

Contact Info

Product

Resources

About