Agents trained by reinforcement learning (RL) often fail to generalize beyond the environment they were trained in, even when presented with new scenarios that seem similar to the training environment. We study the query complexity required to train RL agents that generalize to multiple environments. Intuitively, tractable generalization is only possible when the environments are similar or close in some sense. To capture this, we introduce Weak Proximity, a natural structural condition that requires the environments to have highly similar transition and reward functions and share a policy providing optimal value. Despite such shared structure, we prove that tractable generalization is impossible in the worst case. This holds even when each individual environment can be efficiently solved to obtain an optimal linear policy, and when the agent possesses a generative model. Our lower bound applies to the more complex task of representation learning for the purpose of efficient generalization to multiple environments. On the positive side, we introduce Strong Proximity, a strengthened condition which we prove is sufficient for efficient generalization.
more »
« less
This content will become publicly available on January 1, 2024
InitLight: Initial Model Generation for Traffic Signal Control Using Adversarial Inverse Reinforcement Learning
Due to repetitive trial-and-error style interactions between agents and a fixed traffic environment during the policy learning, existing Reinforcement Learning (RL)-based Traffic Signal Control (TSC) methods greatly suffer from long RL training time and poor adaptability of RL agents to other complex traffic environments. To address these problems, we propose a novel Adversarial Inverse Reinforcement Learning (AIRL)-based pre-training method named InitLight, which enables effective initial model generation for TSC agents. Unlike traditional RL-based TSC approaches that train a large number of agents simultaneously for a specific multi-intersection environment, InitLight pretrains only one single initial model based on multiple single-intersection environments together with
their expert trajectories. Since the reward function learned by InitLight can recover ground-truth TSC rewards for different intersections at optimality, the pre-trained agent can be deployed at intersections of any traffic environments as initial models to accelerate subsequent overall global RL training. Comprehensive experimental results show that, the
initial model generated by InitLight can not only significantly accelerate the convergence with much fewer episodes, but also own superior generalization ability to accommodate various kinds of complex traffic environments.
more »
« less
- Award ID(s):
- 2217104
- NSF-PAR ID:
- 10464779
- Date Published:
- Journal Name:
- IJCAI
- ISSN:
- 1045-0823
- Page Range / eLocation ID:
- 4949-4958
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Connected vehicle (CV) technologies enable data exchange between vehicles and transportation infrastructure. In a CV environment, traffic signal control systems receive CV trajectory data through vehicle-to-infrastructure (V2I) communications to make control decisions. Comparing with existing data collection methods (e.g., from loop-detectors), the CV trajectory data provide much richer information, and therefore have great potentials to improve the system performance by reducing total vehicle delay at signalized intersections. However, this connectivity might also bring cyber security concerns. In this paper, we aim to investigate the security problem of CV-based traffic signal control (CV-TSC) systems. Specifically, we focus on evaluating the impact of falsified data attacks on the system performance. A black-box attack scenario, in which the control logic of a CV-TSC system is unavailable to attackers, is considered. A two-step attack model is constructed. In the first step, the attacker tries to learn the control logic using a surrogate model. Based on the surrogate model, in the second step, the attacker launches falsified data attacks to influence the control systems to make sub-optimal control decisions. In the case study, we apply the attack model to an existing CV-TSC system (i.e., I-SIG) and find intersection delay can be significantly increased. Finally, we discuss some promising defense directions.more » « less
-
Perimeter metering control has long been an active research topic since well-defined relationships between network productivity and usage, that is, network macroscopic fundamental diagrams (MFDs), were shown to be capable of describing regional traffic dynamics. Numerous methods have been proposed to solve perimeter metering control problems, but these generally require knowledge of the MFDs or detailed equations that govern traffic dynamics. Recently, a study applied model-free deep reinforcement learning (Deep-RL) methods to two-region perimeter control and found comparable performances to the model predictive control scheme, particularly when uncertainty exists. However, the proposed methods therein provide very low initial performances during the learning process, which limits their applicability to real life scenarios. Furthermore, the methods may not be scalable to more complicated networks with larger state and action spaces. To combat these issues, this paper proposes to integrate the domain control knowledge (DCK) of congestion dynamics into the agent designs for improved learning and control performances. A novel agent is also developed that builds on the Bang-Bang control policy. Two types of DCK are then presented to provide knowledge-guided exploration strategies for the agents such that they can explore around the most rewarding part of the action spaces. The results from extensive numerical experiments on two- and three-region urban networks show that integrating DCK can (a) effectively improve learning and control performances for Deep-RL agents, (b) enhance the agents’ resilience against various types of environment uncertainties, and (c) mitigate the scalability issue for the agents.more » « less
-
Reinforcement Learning (RL) agents in the real world must satisfy safety constraints in addition to maximizing a reward objective. Model-based RL algorithms hold promise for reducing unsafe real-world actions: they may synthesize policies that obey all constraints using simulated samples from a learned model. However, imperfect models can result in real-world constraint violations even for actions that are predicted to satisfy all constraints. We propose Conservative and Adaptive Penalty (CAP), a model-based safe RL framework that accounts for potential modeling errors by capturing model uncertainty and adaptively exploiting it to balance the reward and the cost objectives. First, CAP inflates predicted costs using an uncertainty-based penalty. Theoretically, we show that policies that satisfy this conservative cost constraint are guaranteed to also be feasible in the true environment. We further show that this guarantees the safety of all intermediate solutions during RL training. Further, CAP adaptively tunes this penalty during training using true cost feedback from the environment. We evaluate this conservative and adaptive penalty-based approach for model-based safe RL extensively on state and image-based environments. Our results demonstrate substantial gains in sample-efficiency while incurring fewer violations than prior safe RL algorithms. Code is available at: https://github.com/Redrew/CAPmore » « less
-
more » « less
Abstract Shared control of mobile robots integrates manual input with auxiliary autonomous controllers to improve the overall system performance. However, prior work that seeks to find the optimal shared control ratio needs an accurate human model, which is usually challenging to obtain. In this study, the authors develop an extended Twin Delayed Deep Deterministic Policy Gradient (DDPG) (TD3X)‐based shared control framework that learns to assist a human operator in teleoperating mobile robots optimally. The robot's states, shared control ratio in the previous time step, and human's control input is used as inputs to the reinforcement learning (RL) agent, which then outputs the optimal shared control ratio between human input and autonomous controllers without knowing the human model. Noisy softmax policies are developed to make the TD3X algorithm feasible under the constraint of a shared control ratio. Furthermore, to accelerate the training process and protect the robot, a navigation demonstration policy and a safety guard are developed. A neural network (NN) structure is developed to maintain the correlation of sensor readings among heterogeneous input data and improve the learning speed. In addition, an extended DAGGER (DAGGERX) human agent is developed for training the RL agent to reduce human workload. Robot simulations and experiments with humans in the loop are conducted. The results show that the DAGGERX human agent can simulate real human inputs in the worst‐case scenarios with a mean square error of 0.0039. Compared to the original TD3 agent, the TD3X‐based shared control system decreased the average collision number from 387.3 to 44.4 in a simplistic environment and 394.2 to 171.2 in a more complex environment. The maximum average return increased from 1043 to 1187 with a faster converge speed in the simplistic environment, while the performance is equally good in the complex environment because of the use of an advanced human agent. In the human subject tests, participants' average perceived workload was significantly lower in shared control than that in exclusively manual control (26.90 vs. 40.07,
p = 0.013).
