Collective action demands that individuals efficiently coordinate how much, where, and when to cooperate. Laboratory experiments have extensively explored the first part of this process, demonstrating that a variety of social-cognitive mechanisms influence how much individuals choose to invest in group efforts. However, experimental research has been unable to shed light on how social cognitive mechanisms contribute to the where and when of collective action. We leverage multi-agent deep reinforcement learning to model how a social-cognitive mechanism--specifically, the intrinsic motivation to achieve a good reputation--steers group behavior toward specific spatial and temporal strategies for collective action in a social dilemma. We also collect behavioral data from groups of human participants challenged with the same dilemma. The model accurately predicts spatial and temporal patterns of group behavior: in this public goods dilemma, the intrinsic motivation for reputation catalyzes the development of a non-territorial, turn-taking strategy to coordinate collective action.

In offline reinforcement learning (RL), we seek to utilize offline data to evaluate (or learn) policies in scenarios where the data are collected from a distribution that substantially differs from that of the target policy to be evaluated. Recent theoretical advances have shown that such sample-efficient offline RL is indeed possible provided certain strong representational conditions hold, else there are lower bounds exhibiting exponential error amplification (in the problem horizon) unless the data collection distribution has only a mild distribution shift relative to the target policy. This work studies these issues from an empirical perspective to gauge how stable offline RL methods are. In particular, our methodology explores these ideas when using features from pre-trained neural networks, in the hope that these representations are powerful enough to permit sample efficient offline RL. Through extensive experiments on a range of tasks, we see that substantial error amplification does occur even when using such pre-trained representations (trained on the same task itself); we find offline RL is stable only under extremely mild distribution shift. The implications of these results, both from a theoretical and an empirical perspective, are that successful offline RL (where we seek to go beyond the low distribution shift regime) requires substantially stronger conditions beyond those which suffice for successful supervised learning.

Despite the rich theoretical foundation of model-based deep reinforcement learning (RL) agents, their effectiveness in real-world robotics-applications is less studied and understood. In this paper, we, therefore, investigate how such agents generalize to real-world autonomous-vehicle control-tasks, where advanced model-free deep RL algorithms fail. In particular, we set up a series of time-lap tasks for an F1TENTH racing robot, equipped with high-dimensional LiDAR sensors, on a set of test tracks with a gradual increase in their complexity. In this continuous-control setting, we show that model-based agents capable of learning in imagination, substantially outperform model-free agents with respect to performance, sample efficiency, successful task completion, and generalization. Moreover, we show that the generalization ability of model-based agents strongly depends on the observation-model choice. Finally, we provide extensive empirical evidence for the effectiveness of model-based agents provided with long enough memory horizons in sim2real tasks.

We present an approach for procedural content generation (PCG), and improving generalization in reinforcement learning (RL) agents, by using adversarial deep RL. Training RL agents for generalization over novel environments is a notoriously difficult task. One popular approach is to procedurally generate different environments to increase the generalizability of the trained agents. Here we deploy an adversarial model with one PCG RL agent (called Generator), and one solving RL agent (called Solver). The benefit is mainly two-fold: Firstly, the Solver achieves better generalization through the generated challenges from the Generator. Secondly, the trained Generator can be used as a creator of novel environments that, together with the Solver, can be shown to be solvable. The Generator receives a reward signal based on the performance of the Solver which encourages the environment design to be challenging but not impossible. To further drive diversity and control of the environment generation, we propose the use of auxiliary inputs for the Generator. Thus, we propose adversarial RL for procedural content generation (ARLPCG), an adversarial approach which procedurally generates previously unseen environments with an auxiliary input as a control variable. Herein we describe this concept in detail and compare it with previous methods showing improved generalization, as well as a new method to create novel environments.

We present a slot-wise, object-based transition model that decomposes a scene into objects, aligns them (with respect to a slot-wise object memory) to maintain a consistent order across time, and predicts how those objects evolve over successive frames. The model is trained end-to-end without supervision using losses at the level of the object-structured representation rather than pixels. Thanks to its alignment module, the model deals properly with two issues that are not handled satisfactorily by other transition models, namely object persistence and object identity. We show that the combination of an object-level loss and correct object alignment over time enables the model to outperform a state-of-the-art baseline, and allows it to deal well with object occlusion and re-appearance in partially observable environments.

This letter compares the performance of four different, popular simulation environments for robotics and reinforcement learning (RL) through a series of benchmarks. The benchmarked scenarios are designed carefully with current industrial applications in mind. Given the need to run simulations as fast as possible to reduce the real-world training time of the RL agents, the comparison includes not only different simulation environments but also different hardware configurations, ranging from an entry-level notebook up to a dual CPU high performance server. We show that the chosen simulation environments benefit the most from single core performance. Yet, using a multi core system, multiple simulations could be run in parallel to increase the performance.

We present a new practical framework based on deep reinforcement learning and decision-time planning for real-world vehicle repositioning on ride-hailing (a type of mobility-on-demand, MoD) platforms. Our approach learns the spatiotemporal state-value function using a batch training algorithm with deep value networks. The optimal repositioning action is generated on-demand through value-based policy search, which combines planning and bootstrapping with the value networks. For the large-fleet problems, we develop several algorithmic features that we incorporate into our framework and that we demonstrate to induce coordination among the algorithmically-guided vehicles. We benchmark our algorithm with baselines in a ride-hailing simulation environment to demonstrate its superiority in improving income efficiency meausred by income-per-hour. We have also designed and run a real-world experiment program with regular drivers on a major ride-hailing platform. We have observed significantly positive results on key metrics comparing our method with experienced drivers who performed idle-time repositioning based on their own expertise.

Unlike theoretical distributed learning (DL), DL over wireless edge networks faces the inherent dynamics/uncertainty of wireless connections and edge nodes, making DL less efficient or even inapplicable under the highly dynamic wireless edge networks (e.g., using mmW interfaces). This article addresses these problems by leveraging recent advances in coded computing and the deep dueling neural network architecture. By introducing coded structures/redundancy, a distributed learning task can be completed without waiting for straggling nodes. Unlike conventional coded computing that only optimizes the code structure, coded distributed learning over the wireless edge also requires to optimize the selection/scheduling of wireless edge nodes with heterogeneous connections, computing capability, and straggling effects. However, even neglecting the aforementioned dynamics/uncertainty, the resulting joint optimization of coding and scheduling to minimize the distributed learning time turns out to be NP-hard. To tackle this and to account for the dynamics and uncertainty of wireless connections and edge nodes, we reformulate the problem as a Markov Decision Process and then design a novel deep reinforcement learning algorithm that employs the deep dueling neural network architecture to find the jointly optimal coding scheme and the best set of edge nodes for different learning tasks without explicit information about the wireless environment and edge nodes' straggling parameters. Simulations show that the proposed framework reduces the average learning delay in wireless edge computing up to 66% compared with other DL approaches. The jointly optimal framework in this article is also applicable to any distributed learning scheme with heterogeneous and uncertain computing nodes.

Recently, I was at a party in San Francisco when a man approached me and introduced himself as the founder of a small artificial intelligence (AI) start-up. As soon as the founder figured out that I was a technology writer for The New York Times, he launched into a pitch for his company, which he said was trying to revolutionise the manufacturing sector using a new AI technique called "deep reinforcement learning". The founder explained that his company's AI could run millions of virtual simulations for any given factory, eventually arriving at the exact sequence of processes that would allow it to produce goods most efficiently. This AI, he said, would allow factories to replace entire teams of human production planners, along with most of the outdated software those people relied on. "We call it the Boomer Remover," he said.

Inverse reinforcement learning (IRL) is the task of finding a reward function that generates a desired optimal policy for a given Markov Decision Process (MDP). This paper develops an information-theoretic lower bound for the sample complexity of the finite state, finite action IRL problem. A geometric construction of $\beta$-strict separable IRL problems using spherical codes is considered. Properties of the ensemble size as well as the Kullback-Leibler divergence between the generated trajectories are derived. The resulting ensemble is then used along with Fano's inequality to derive a sample complexity lower bound of $O(n \log n)$, where $n$ is the number of states in the MDP.