Dyna-style reinforcement learning (RL) agents improve sample efficiency over model-free RL agents by updating the value function with simulated experience generated by an environment model. However, it is often difficult to learn accurate models of environment dynamics, and even small errors may result in failure of Dyna agents. In this paper, we highlight that one potential cause of that failure is bootstrapping off of the values of simulated states, and introduce a new Dyna algorithm to avoid this failure. We discuss a design space of Dyna algorithms, based on using successor or predecessor models---simulating forwards or backwards---and using one-step or multi-step updates. Three of the variants have been explored, but surprisingly the fourth variant has not: using predecessor models with multi-step updates. We present the \emph{Hallucinated Value Hypothesis} (HVH): updating the values of real states towards values of simulated states can result in misleading action values which adversely affect the control policy. We discuss and evaluate all four variants of Dyna amongst which three update real states toward simulated states --- so potentially toward hallucinated values --- and our proposed approach, which does not. The experimental results provide evidence for the HVH, and suggest that using predecessor models with multi-step updates is a fruitful direction toward developing Dyna algorithms that are more robust to model error.

The Model AI Assignments session seeks to gather and disseminate the best assignment designs of the Artificial Intelligence (AI) Education community. Recognizing that assignments form the core of student learning experience, we here present abstracts of six AI assignments from the 2023 session that are easily adoptable, playfully engaging, and flexible for a variety of instructor needs. Assignment specifications and supporting resources may be found at http://modelai.gettysburg.edu .

In model-based reinforcement learning, planning with an imperfect model of the environment has the potential to harm learning progress. But even when a model is imperfect, it may still contain information that is useful for planning. In this paper, we investigate the idea of using an imperfect model selectively. The agent should plan in parts of the state space where the model would be helpful but refrain from using the model where it would be harmful. An effective selective planning mechanism requires estimating predictive uncertainty, which arises out of aleatoric uncertainty, parameter uncertainty, and model inadequacy, among other sources. Prior work has focused on parameter uncertainty for selective planning. In this work, we emphasize the importance of model inadequacy. We show that heteroscedastic regression can signal predictive uncertainty arising from model inadequacy that is complementary to that which is detected by methods designed for parameter uncertainty, indicating that considering both parameter uncertainty and model inadequacy may be a more promising direction for effective selective planning than either in isolation.

In model-based reinforcement learning it is typical to decouple the problems of learning the dynamics model and learning the reward function. However, when the dynamics model is flawed, it may generate erroneous states that would never occur in the true environment. It is not clear a priori what value the reward function should assign to such states. This paper presents a novel error bound that accounts for the reward model’s behavior in states sampled from the model. This bound is used to extend the existing Hallucinated DAgger-MC algorithm, which offers theoretical performance guarantees in deterministic MDPs that do not assume a perfect model can be learned. Empirically, this approach to reward learning can yield dramatic improvements in control performance when the dynamics model is flawed.

The Arcade Learning Environment (ALE) is an evaluation platform that poses the challenge of building AI agents with general competency across dozens of Atari 2600 games. It supports a variety of different problem settings and it has been receiving increasing attention from the scientific community. In this paper we take a big picture look at how the ALE is being used by the research community. We focus on how diverse the evaluation methodologies in the ALE have become and we highlight some key concerns when evaluating agents in this platform. We use this discussion to present what we consider to be the best practices for future evaluations in the ALE. To further the progress in the field, we also introduce a new version of the ALE that supports multiple game modes and provides a form of stochasticity we call sticky actions.

The Arcade Learning Environment (ALE) is an evaluation platform that poses the challenge of building AI agents with general competency across dozens of Atari 2600 games. It supports a variety of different problem settings and it has been receiving increasing attention from the scientific community, leading to some high-profile success stories such as the much publicized Deep Q-Networks (DQN). In this article we take a big picture look at how the ALE is being used by the research community. We show how diverse the evaluation methodologies in the ALE have become with time, and highlight some key concerns when evaluating agents in the ALE. We use this discussion to present some methodological best practices and provide new benchmark results using these best practices. To further the progress in the field, we introduce a new version of the ALE that supports multiple game modes and provides a form of stochasticity we call sticky actions. We conclude this big picture look by revisiting challenges posed when the ALE was introduced, summarizing the state-of-the-art in various problems and highlighting problems that remain open.

When an agent cannot represent a perfectly accurate model of its environment's dynamics, model-based reinforcement learning (MBRL) can fail catastrophically. Planning involves composing the predictions of the model; when flawed predictions are composed, even minor errors can compound and render the model useless for planning. Hallucinated Replay (Talvitie 2014) trains the model to "correct" itself when it produces errors, substantially improving MBRL with flawed models. This paper theoretically analyzes this approach, illuminates settings in which it is likely to be effective or ineffective, and presents a novel error bound, showing that a model's ability to self-correct is more tightly related to MBRL performance than one-step prediction error. These results inspire an MBRL algorithm for deterministic MDPs with performance guarantees that are robust to model class limitations.

Much of the focus on finding good representations in reinforcement learning has been on learning complex non-linear predictors of value. Policy gradient algorithms, which directly represent the policy, often need fewer parameters to learn good policies. However, they typically employ a fixed parametric representation that may not be sufficient for complex domains. This paper introduces the Policy Tree algorithm, which can learn an adaptive representation of policy in the form of a decision tree over different instantiations of a base policy. Policy gradient is used both to optimize the parameters and to grow the tree by choosing splits that enable the maximum local increase in the expected return of the policy. Experiments show that this algorithm can choose genuinely helpful splits and significantly improve upon the commonly used linear Gibbs softmax policy, which we choose as our base policy.

Monte-Carlo planning has been proven successful in manysequential decision-making settings, but it suffers from poorexploration when the rewards are sparse. In this paper, weimprove exploration in UCT by generalizing across similarstates using a given distance metric. We show that this algorithm,like UCT, converges asymptotically to the optimalaction. When the state space does not have a natural distancemetric, we show how we can learn a local manifold from thetransition graph of states in the near future. to obtain a distancemetric. On domains inspired by video games, empiricalevidence shows that our algorithm is more sample efficientthan UCT, particularly when rewards are sparse.

While model-based reinforcement learning is often studied under the assumption that a fully accurate model is contained within the model class, this is rarely true in practice. When the model class may be fundamentally limited, it can be difficult to obtain theoretical guarantees. Under some conditions the DAgger algorithm promises a policy nearly as good as the plan obtained from the most accurate model in the class, but only if the planning algorithm is near-optimal, which is also rarely the case in complex problems. This paper explores the interaction between DAgger and Monte Carlo planning, specifically showing that DAgger may perform poorly when coupled with a sub-optimal planner. A novel variation of DAgger specifically for use with Monte Carlo planning is derived and is shown to behave far better in some cases where DAgger fails.

Context Tree Weighting (CTW) is a powerful probabilistic sequence prediction technique that efficiently performs Bayesian model averaging over the class of all prediction suffix trees of bounded depth. In this paper we show how to generalize this technique to the class of K-skip prediction suffix trees. Contrary to regular prediction suffix trees, K-skip prediction suffix trees are permitted to ignore up to K contiguous portions of the context. This allows for significant improvements in predictive accuracy when irrelevant variables are present, a case which often occurs within record-aligned data and images. We provide a regret-based analysis of our approach, and empirically evaluate it on the Calgary corpus and a set of Atari 2600 screen prediction tasks.

This paper introduces timeline trees, which are partial models of partially observable environments. Timeline trees are given some specific predictions to make and learn a decision tree over history. The main idea of timeline trees is to use temporally abstract features to identify and split on features of key events, spread arbitrarily far apart in the past (whereas previous decision-tree-based methods have been limited to a finite suffix of history). Experiments demonstrate that timeline trees can learn to make high quality predictions in complex, partially observable environments with high-dimensional observations (e.g. an arcade game).

When faced with the problem of learning a model of a high-dimensional environment, a common approach is to limit the model to make only a restricted set of predictions, thereby simplifying the learning problem. These partial models may be directly useful for making decisions or may be combined together to form a more complete, structured model. However, in partially observable (non-Markov) environments, standard model-learning methods learn generative models, i.e. models that provide a probability distribution over all possible futures (such as POMDPs). It is not straightforward to restrict such models to make only certain predictions, and doing so does not always simplify the learning problem. In this paper we present prediction profile models: non-generative partial models for partially observable systems that make only a given set of predictions, and are therefore far simpler than generative models in some cases. We formalize the problem of learning a prediction profile model as a transformation of the original model-learning problem, and show empirically that one can learn prediction profile models that make a small set of important predictions even in systems that are too complex for standard generative models.

We present a novel mathematical formalism for the idea of a local model,'' a model of a potentially complex dynamical system that makes only certain predictions in only certain situations. As a result of its restricted responsibilities, a local model may be far simpler than a complete model of the system. We then show how one might combine several local models to produce a more detailed model. We demonstrate our ability to learn a collection of local models on a large-scale example and do a preliminary empirical comparison of learning a collection of local models and some other model learning methods."