Learning tabula rasa, that is without any prior knowledge, is the prevalent workflow in reinforcement learning (RL) research. However, RL systems, when applied to large-scale settings, rarely operate tabula rasa. Such large-scale systems undergo multiple design or algorithmic changes during their development cycle and use ad hoc approaches for incorporating these changes without re-training from scratch, which would have been prohibitively expensive. Additionally, the inefficiency of deep RL typically excludes researchers without access to industrial-scale resources from tackling computationally-demanding problems. To address these issues, we present reincarnating RL as an alternative workflow or class of problem settings, where prior computational work (e.g., learned policies) is reused or transferred between design iterations of an RL agent, or from one RL agent to another. As a step towards enabling reincarnating RL from any agent to any other agent, we focus on the specific setting of efficiently transferring an existing sub-optimal policy to a standalone value-based RL agent. We find that existing approaches fail in this setting and propose a simple algorithm to address their limitations. Equipped with this algorithm, we demonstrate reincarnating RL's gains over tabula rasa RL on Atari 2600 games, a challenging locomotion task, and the real-world problem of navigating stratospheric balloons. Overall, this work argues for an alternative approach to RL research, which we believe could significantly improve real-world RL adoption and help democratize it further. Open-sourced code and trained agents at https://agarwl.github.io/reincarnating_rl.

In the following article I discuss how to generate emergent behavior in AI characters using Unreal Engine, Reinforcement Learning, and the free machine learning plugin MindMaker. The aim is that the interested reader can use this as a guide for creating emergent behavior in their own game project or embodied AI character. Emergent behavior refers to behaviors that are not pre-programmed but develop organically in response to some environmental stimuli. Emergent behavior is common to many if not all forms of life, being a function of evolution itself. It is also more recently a feature of embodied artificial agents. When one employs emergent behavior methods, one does not rigidly program specific actions for the AI, but instead allows them to "evolve" through some adaptive algorithm such as genetic programming, reinforcement learning, or Monte Carlo methods.

Our world is full of asymmetries. Gravity and wind can make reaching a place easier than coming back. Social artifacts such as genealogy charts and citation graphs are inherently directed. In reinforcement learning and control, optimal goal-reaching strategies are rarely reversible (symmetrical). Distance functions supported on these asymmetrical structures are called quasimetrics. Despite their common appearance, little research has been done on the learning of quasimetrics. Our theoretical analysis reveals that a common class of learning algorithms, including unconstrained multilayer perceptrons (MLPs), provably fails to learn a quasimetric consistent with training data. In contrast, our proposed Poisson Quasimetric Embedding (PQE) is the first quasimetric learning formulation that both is learnable with gradient-based optimization and enjoys strong performance guarantees. Experiments on random graphs, social graphs, and offline Q-learning demonstrate its effectiveness over many common baselines.

Reinforcement Learning (RL) is a general framework concerned with an agent that seeks to maximize rewards in an environment. The learning typically happens through trial and error using explorative methods, such as epsilon-greedy. There are two approaches, model-based and model-free reinforcement learning, that show concrete results in several disciplines. Model-based RL learns a model of the environment for learning the policy while model-free approaches are fully explorative and exploitative without considering the underlying environment dynamics. Model-free RL works conceptually well in simulated environments, and empirical evidence suggests that trial and error lead to a near-optimal behavior with enough training. On the other hand, model-based RL aims to be sample efficient, and studies show that it requires far less training in the real environment for learning a good policy. A significant challenge with RL is that it relies on a well-defined reward function to work well for complex environments and such a reward function is challenging to define. Goal-Directed RL is an alternative method that learns an intrinsic reward function with emphasis on a few explored trajectories that reveals the path to the goal state. This paper introduces a novel reinforcement learning algorithm for predicting the distance between two states in a Markov Decision Process. The learned distance function works as an intrinsic reward that fuels the agent's learning. Using the distance-metric as a reward, we show that the algorithm performs comparably to model-free RL while having significantly better sample-efficiently in several test environments.

Multi-Agent Reinforcement Learning (MARL) -- where multiple agents learn to interact in a shared dynamic environment -- permeates across a wide range of critical applications. While there has been substantial progress on understanding the global convergence of policy optimization methods in single-agent RL, designing and analysis of efficient policy optimization algorithms in the MARL setting present significant challenges, which unfortunately, remain highly inadequately addressed by existing theory. In this paper, we focus on the most basic setting of competitive multi-agent RL, namely two-player zero-sum Markov games, and study equilibrium finding algorithms in both the infinite-horizon discounted setting and the finite-horizon episodic setting. We propose a single-loop policy optimization method with symmetric updates from both agents, where the policy is updated via the entropy-regularized optimistic multiplicative weights update (OMWU) method and the value is updated on a slower timescale. We show that, in the full-information tabular setting, the proposed method achieves a finite-time last-iterate linear convergence to the quantal response equilibrium of the regularized problem, which translates to a sublinear last-iterate convergence to the Nash equilibrium by controlling the amount of regularization. Our convergence results improve upon the best known iteration complexities, and lead to a better understanding of policy optimization in competitive Markov games.

Traffic signal control (TSC) is a high-stakes domain that is growing in importance as traffic volume grows globally. An increasing number of works are applying reinforcement learning (RL) to TSC; RL can draw on an abundance of traffic data to improve signalling efficiency. However, RL-based signal controllers have never been deployed. In this work, we provide the first review of challenges that must be addressed before RL can be deployed for TSC. We focus on four challenges involving (1) uncertainty in detection, (2) reliability of communications, (3) compliance and interpretability, and (4) heterogeneous road users. We show that the literature on RL-based TSC has made some progress towards addressing each challenge. However, more work should take a systems thinking approach that considers the impacts of other pipeline components on RL.

We propose a Proportional Integral Derivative (PID) controller-based coaching scheme to expedite reinforcement learning (RL).

In the coming years and decades, autonomous vehicles (AVs) will become increasingly prevalent, offering new opportunities for safer and more convenient travel and potentially smarter traffic control methods exploiting automation and connectivity. Car following is a prime function in autonomous driving. Car following based on reinforcement learning has received attention in recent years with the goal of learning and achieving performance levels comparable to humans. However, most existing RL methods model car following as a unilateral problem, sensing only the vehicle ahead. Recent literature, however, Wang and Horn [16] has shown that bilateral car following that considers the vehicle ahead and the vehicle behind exhibits better system stability. In this paper we hypothesize that this bilateral car following can be learned using RL, while learning other goals such as efficiency maximisation, jerk minimization, and safety rewards leading to a learned model that outperforms human driving. We propose and introduce a Deep Reinforcement Learning (DRL) framework for car following control by integrating bilateral information into both state and reward function based on the bilateral control model (BCM) for car following control. Furthermore, we use a decentralized multi-agent reinforcement learning framework to generate the corresponding control action for each agent. Our simulation results demonstrate that our learned policy is better than the human driving policy in terms of (a) inter-vehicle headways, (b) average speed, (c) jerk, (d) Time to Collision (TTC) and (e) string stability.

Recent efforts to learn reward functions from human feedback have tended to use deep neural networks, whose lack of transparency hampers our ability to explain agent behaviour or verify alignment. We explore the merits of learning intrinsically interpretable tree models instead. We develop a recently proposed method for learning reward trees from preference labels, and show it to be broadly competitive with neural networks on challenging high-dimensional tasks, with good robustness to limited or corrupted data. Having found that reward tree learning can be done effectively in complex settings, we then consider why it should be used, demonstrating that the interpretable reward structure gives significant scope for traceability, verification and explanation.

Robust reinforcement learning (RL) considers the problem of learning policies that perform well in the worst case among a set of possible environment parameter values. In real-world environments, choosing the set of possible values for robust RL can be a difficult task. When that set is specified too narrowly, the agent will be left vulnerable to reasonable parameter values unaccounted for. When specified too broadly, the agent will be too cautious. In this paper, we propose Feasible Adversarial Robust RL (FARR), a novel problem formulation and objective for automatically determining the set of environment parameter values over which to be robust. FARR implicitly defines the set of feasible parameter values as those on which an agent could achieve a benchmark reward given enough training resources. By formulating this problem as a two-player zero-sum game, optimizing the FARR objective jointly produces an adversarial distribution over parameter values with feasible support and a policy robust over this feasible parameter set. We demonstrate that approximate Nash equilibria for this objective can be found using a variation of the PSRO algorithm. Furthermore, we show that an optimal agent trained with FARR is more robust to feasible adversarial parameter selection than with existing minimax, domain-randomization, and regret objectives in a parameterized gridworld and three MuJoCo control environments.