Developing reinforcement learning algorithms that satisfy safety constraints is becoming increasingly important in real-world applications. In multi-agent reinforcement learning (MARL) settings, policy optimisation with safety awareness is particularly challenging because each individual agent has to not only meet its own safety constraints, but also consider those of others so that their joint behaviour can be guaranteed safe. Despite its importance, the problem of safe multi-agent learning has not been rigorously studied; very few solutions have been proposed, nor a sharable testing environment or benchmarks. To fill these gaps, in this work, we formulate the safe MARL problem as a constrained Markov game and solve it with policy optimisation methods. Our solutions -- Multi-Agent Constrained Policy Optimisation (MACPO) and MAPPO-Lagrangian -- leverage the theories from both constrained policy optimisation and multi-agent trust region learning. Crucially, our methods enjoy theoretical guarantees of both monotonic improvement in reward and satisfaction of safety constraints at every iteration. To examine the effectiveness of our methods, we develop the benchmark suite of Safe Multi-Agent MuJoCo that involves a variety of MARL baselines. Experimental results justify that MACPO/MAPPO-Lagrangian can consistently satisfy safety constraints, meanwhile achieving comparable performance to strong baselines.

When it comes to games such as chess or Go, artificial intelligence (AI) programs have far surpassed the best players in the world. These "superhuman" AIs are unmatched competitors, but perhaps harder than competing against humans is collaborating with them. Can the same technology get along with people? In a new study, MIT Lincoln Laboratory researchers sought to find out how well humans could play the cooperative card game Hanabi with an advanced AI model trained to excel at playing with teammates it had never met before. In single-blind experiments, participants played two series of the game: One with the AI agent as their teammate, and the other with a rule-based agent, a bot manually programmed to play in a predefined way.

Learning in multi-agent environments is difficult due to the non-stationarity introduced by an opponent's or partner's changing behaviors. Instead of reactively adapting to the other agent's (opponent or partner) behavior, we propose an algorithm to proactively influence the other agent's strategy to stabilize -- which can restrain the non-stationarity caused by the other agent. We learn a low-dimensional latent representation of the other agent's strategy and the dynamics of how the latent strategy evolves with respect to our robot's behavior. With this learned dynamics model, we can define an unsupervised stability reward to train our robot to deliberately influence the other agent to stabilize towards a single strategy. We demonstrate the effectiveness of stabilizing in improving efficiency of maximizing the task reward in a variety of simulated environments, including autonomous driving, emergent communication, and robotic manipulation. We show qualitative results on our website: https://sites.google.com/view/stable-marl/.

Frequent lane changes during congestion at freeway bottlenecks such as merge and weaving areas further reduce roadway capacity. The emergence of deep reinforcement learning (RL) and connected and automated vehicle technology provides a possible solution to improve mobility and energy efficiency at freeway bottlenecks through cooperative lane changing. Deep RL is a collection of machine-learning methods that enables an agent to improve its performance by learning from the environment. In this study, a decentralized cooperative lane-changing controller was developed using proximal policy optimization by adopting a multi-agent deep RL paradigm. In the decentralized control strategy, policy learning and action reward are evaluated locally, with each agent (vehicle) getting access to global state information. Multi-agent deep RL requires lower computational resources and is more scalable than single-agent deep RL, making it a powerful tool for time-sensitive applications such as cooperative lane changing. The results of this study show that cooperative lane changing enabled by multi-agent deep RL yields superior performance to human drivers in term of traffic throughput, vehicle speed, number of stops per vehicle, vehicle fuel efficiency, and emissions. The trained RL policy is transferable and can be generalized to uncongested, moderately congested, and extremely congested traffic conditions.

In the absence of prescribed coordination, it is often necessary for individual agents to synthesize their own plans, taking into account not only their own capabilities and beliefs about the world but also their beliefs about other agents, including what each of the agents will come to believe as the consequence of the actions of others. To illustrate, consider the scenario where Larry and Moe meet on a regular basis at the local diner to swap the latest gossip. Larry has come to know that Nancy (Larry's daughter) has just received a major promotion in her job, but unbeknownst to him, Moe has already learned this bit of information through the grapevine. Before they speak, both believe Nancy is getting a promotion, Larry believes Moe is unaware of this (and consequently wishes to share the news), and Moe assumes Larry must already be aware of the promotion but is unaware of Moe's own knowledge of the situation. Very quickly we can see how the nesting of (potentially incorrect) belief can be a complicated and interesting setting to model. In this paper, we examine the problem of synthesizing plans in such settings. In particular, given a finite set of agents, each with: (1) (possibly incomplete and incorrect) beliefs about the world and about the beliefs of other agents; and (2) differing capabilities including the ability to perform actions whose outcomes are unknown to other agents; we are interested in synthesizing a plan to achieve a goal condition. Planning is at the belief level and as such, while we consider the execution of actions that can change the state of the world (ontic actions) as well as an agent's state of knowledge or belief (epistemic or more accurately doxastic actions, including communication actions), all outcomes are with respect to belief.

Deep reinforcement learning (RL) agents may successfully generalize to new settings if trained on an appropriately diverse set of environment and task configurations. Unsupervised Environment Design (UED) is a promising self-supervised RL paradigm, wherein the free parameters of an underspecified environment are automatically adapted during training to the agent's capabilities, leading to the emergence of diverse training environments. Here, we cast Prioritized Level Replay (PLR), an empirically successful but theoretically unmotivated method that selectively samples randomly-generated training levels, as UED. We argue that by curating completely random levels, PLR, too, can generate novel and complex levels for effective training. This insight reveals a natural class of UED methods we call Dual Curriculum Design (DCD). Crucially, DCD includes both PLR and a popular UED algorithm, PAIRED, as special cases and inherits similar theoretical guarantees. This connection allows us to develop novel theory for PLR, providing a version with a robustness guarantee at Nash equilibria. Furthermore, our theory suggests a highly counterintuitive improvement to PLR: by stopping the agent from updating its policy on uncurated levels (training on less data), we can improve the convergence to Nash equilibria. Indeed, our experiments confirm that our new method, PLR$^{\perp}$, obtains better results on a suite of out-of-distribution, zero-shot transfer tasks, in addition to demonstrating that PLR$^{\perp}$ improves the performance of PAIRED, from which it inherited its theoretical framework.

AI systems have seen dramatic advancement in recent years, bringing many applications that pervade our everyday life. However, we are still mostly seeing instances of narrow AI: many of these recent developments are typically focused on a very limited set of competencies and goals, e.g., image interpretation, natural language processing, classification, prediction, and many others. Moreover, while these successes can be accredited to improved algorithms and techniques, they are also tightly linked to the availability of huge datasets and computational power. State-of-the-art AI still lacks many capabilities that would naturally be included in a notion of (human) intelligence. We argue that a better study of the mechanisms that allow humans to have these capabilities can help us understand how to imbue AI systems with these competencies. We focus especially on D. Kahneman's theory of thinking fast and slow, and we propose a multi-agent AI architecture where incoming problems are solved by either system 1 (or "fast") agents, that react by exploiting only past experience, or by system 2 (or "slow") agents, that are deliberately activated when there is the need to reason and search for optimal solutions beyond what is expected from the system 1 agent. Both kinds of agents are supported by a model of the world, containing domain knowledge about the environment, and a model of "self", containing information about past actions of the system and solvers' skills.

Simulation is used extensively in autonomous systems, particularly in robotic manipulation. By far, the most common approach is to train a controller in simulation, and then use it as an initial starting point for the real system. We demonstrate how to learn simultaneously from both simulation and interaction with the real environment. We propose an algorithm for balancing the large number of samples from the high throughput but less accurate simulation and the low-throughput, high-fidelity and costly samples from the real environment. We achieve that by maintaining a replay buffer for each environment the agent interacts with. We analyze such multi-environment interaction theoretically, and provide convergence properties, through a novel theoretical replay buffer analysis. We demonstrate the efficacy of our method on a sim-to-real environment.

Particle swarm optimization (PSO) is one of the bio-inspired algorithms and it is a simple one to search for an optimal solution in the solution space. It is different from other optimization algorithms in such a way that only the objective function is needed and it is not dependent on the gradient or any differential form of the objective. It also has very few hyperparameters. In this tutorial, you will learn the rationale of PSO and its algorithm with an example. Particle Swarm Optimization was proposed by Kennedy and Eberhart in 1995.

Attempts to integrate ToM in opponent modeling has profound cognitive science origin de2014theory; baker2017rational. LOLA in foerster2018learning learns the best response to evolving opponents. Yet, opponents/partners' real-time believes are not considered into policy. However, I-POMDP requires extensive sampling to approximate the nested integration over the belief space, action space and observation space, limiting its scalability. The Bayesian action decoder (BAD)-MDP proposed by Foerster et al.