Learning from Demonstration for Shaping through Inverse Reinforcement Learning

Robots had a great impact on the manufacturing industry ever since the early seventies when companies such as KUKA and ABB started deploying their first industrial robots. These robots merely performed very specific tasks in specific ways within well-defined environments. Still, they proved to be very useful as they could exceed human performance at these tasks. However, in order to enable robots to enter our daily life, they need to become more versatile and need to operate in much less structured environments. This thesis is partly devoted to stretching these limitations by means of learning, namely imitation learning (IL) and inverse reinforcement learning (IRL). Reinforcement learning (RL) is a powerful approach to enable robots to solve a task in an unknown environment. The practitioner describes a desired behavior by specifying a reward function and the robot autonomously interacts with the environment in order to find a control policy that generates high accumulated reward. However, RL is not suitable for teaching new tasks by non-experts because specifying appropriate reward functions can be difficult. Demonstrating the desired behavior is often easier for non-experts. Imitation learning can be used in order to enable the robot to reproduce the demonstrations. However, without explicitly inferring and modeling the intentions of the demonstrations, it can become difficult to solve the task for unseen situations. Inverse reinforcement learning (IRL) therefore aims to infer a reward function from the demonstrations, such that optimizing this reward function yields the desired behavior even for different situations. This thesis introduces a unifying approach to solve the inverse reinforcement learning problem in the same way as the reinforcement learning problem. This is achieved by framing both problems as information projection problems, i.e., we strive to minimize the relative entropy between a probabilistic model of the robot behavior and a given desired distribution. Furthermore, a trust region on the robot behavior is used to stabilize the optimization. For inverse reinforcement learning, the desired distribution is implicitly given by the expert demonstrations. The resulting optimization can be efficiently solved using state-of-the-art reinforcement learning methods. For reinforcement learning, the log-likelihood of the desired distribution is given by the reward function. The resulting optimization problem corresponds to a standard reinforcement learning formulation, except for an additional objective of maximizing the entropy of the robot behavior. This entropy objective adds little overhead to the optimization, but can lead to better exploration and more diversified policies. Trust-region I-projections are not only useful for training robots, but can also be applied to other machine learning problems. I-projections are typically used for variational inference, in order to approximate an intractable distribution by a simpler model. However, the resulting optimization problems are usually optimized based on stochastic gradient descent which often suffers from high variance in the gradient estimates. As trust-region I-projections where shown to be effective for reinforcement learning and inverse reinforcement learning, this thesis also explores their use for variational inference. More specifically, trust-region I-projections are investigated for the problem of approximating an intractable distribution by a Gaussian mixture model (GMM) with an adaptive number of components. GMMs are highly desirable for variational inference because they can yield arbitrary accurate approximations while inference from GMMs is still relatively cheap. In order to make learning the GMM feasible, we derive a lower bound that enables us to decompose the objective function. The optimization can then be performed by iteratively updating individual components using a technique from reinforcement learning. The resulting method is capable of learning approximations of significantly higher quality than existing variational inference methods. Due to the similarity of the underlying optimization problems, the insights gained from our variational inference method are also useful for IL and IRL. Namely, a similar lower bound can be applied also for the I-projection formulation of imitation learning. However, whereas for variational inference the lower bound serves to decompose the objective function, for imitation learning it allows us to provide a reward signal to the robot that does not depend on its behavior. Compared to reward functions that are relative to the current behavior of the robot---which are typical for popular adversarial methods---behavior-independent reward functions have the advantages that we can show convergence even for greedy optimization. Furthermore, behavior-independent reward functions solve the inverse reinforcement learning problem, thereby closing the gap between imitation learning and IRL. However, algorithms derived from our non-adversarial formulation are actually very similar to existing AIL methods, and we can even show that adversarial inverse reinforcement learning (AIRL) is indeed an instance of our formulation. AIRL was derived from an adversarial formulation, and we point out several problems of that derivation. In contrast, we show that AIRL can be straightforwardly derived from out non-adversarial formulation. Furthermore, we demonstrate that the non-adversarial formulation can be also used to derive novel algorithms by presenting a non-adversarial method for offline imitation learning.

Animals integrate knowledge about how the state of the environment evolves to choose actions that maximise reward. Such goal-directed behaviour - or model-based (MB) reinforcement learning (RL) - can flexibly adapt choice to changes, being thus distinct from simpler habitual - or model-free (MF) RL - strategies. Previous inactivation and neuroimaging work implicates prefrontal cortex (PFC) and the caudate striatal region in MB-RL; however, details are scarce about its implementation at the single-neuron level. Here, we recorded from two PFC regions – the dorsal anterior cingulate cortex (ACC) and dorsolateral PFC (DLPFC), and two striatal regions, caudate and putamen – while two rhesus macaques performed a sequential decision-making (two-step) task in which MB-RL involves knowledge about the statistics of reward and state transitions. All four regions, but particularly the ACC, encoded the rewards received and tracked the probabilistic state transitions that occurred. However, ACC (and to a lesser extent caudate) encoded the key variables of the task - namely the interaction between reward, transition and choice – which underlies MB decision-making. ACC and caudate neurons also encoded MB-derived estimates of choice values. Moreover, caudate value estimates of the choice options flipped when a rare transition occurred, demonstrating value update based on structural knowledge of the task. The striatal regions were unique (relative to PFC) in encoding the current and previous rewards with opposing polarities, reminiscent of dopaminergic neurons, and indicative of a MF prediction error. Our findings provide a deeper understanding of selective and temporally dissociable neural mechanisms underlying goal-directed behaviour.

Reinforcement learning with reward shaping is a well-established but often computationally expensive approach to multiagent problems. Agent partitioning can assist in this computational complexity by treating each partition of agents as an independent problem. We introduce a novel agent partitioning approach called Reward/Utility-Based Impact (RUBI). RUBI finds an effective partitioning of agents while requiring no prior domain knowledge, provides better performance by discovering a non-trivial agent partitioning, and leads to faster simulations. We test RUBI in the Air Traffic Flow Management Problem, where there are simultaneously tens of thousands of aircraft affecting the system and no intuitive similarity metric between agents. When partitioning with RUBI in the ATFMP, there is a 37% increase in performance, with a 510x speed up per simulation step over non-partitioning approaches.

Arbitration between model-based (MB) and model-free (MF) reinforcement learning (RL) is key feature of human reinforcement learning. The computational model of arbitration control has been demonstrated to outperform conventional reinforcement learning algorithm, in terms of not only behavioral data but also neural signals. However, this arbitration process does not take full account of contextual changes in environment during learning. By incorporating a Dirichlet process Gaussian mixture model into the arbitration process, we propose a meta-controller for RL that quickly adapts to contextual changes of environment. The proposed model performs better than a conventional model-free RL, model-based RL, and arbitration model.

Foreign students’ process of learning Turkish has been a neglected research topic in studies conducted on teaching Turkish to foreigners. This study tackles the process foreign students go through while learning Turkish within the dimension of learning domains. The aim of this study is to obtain information about the Turkish learning process of foreign students who come to Turkey for higher education. This research is a qualitative study. Criterion sampling was used in determining the sample of the study. The criterion of this research was Tanzanian students who have learnt basic Turkish at Gazi TOMER during the 2013-2014 academic year. Data gathering was conducted through document analysis and focus group interviews. The data gathered through semi-structured interviews were analyzed by inductive coding and descriptive analysis method. According to the investigation of the students’ perceptions within the dimension of learning domains, the study revealed that the domain the students had most difficulties was listening, while the least difficulty was observed in reading.

In this paper, we define a novel inverse reinforcement learning (IRL) problem where the demonstrations are multi-intention, i.e., collected from multi-intention experts, unlabeled, i.e., without intention labels, and partially overlapping, i.e., shared between multiple intentions. In the presence of overlapping demonstrations, current IRL methods, developed to handle multi-intention and unlabeled demonstrations, cannot successfully learn the underlying reward functions. To solve this limitation, we propose a novel clustering-based approach to disentangle the observed demonstrations and experimentally validate its advantages. Traditional clustering-based approaches to multi-intention IRL, which are developed on the basis of model-based Reinforcement Learning (RL), formulate the problem using parametric density estimation. However, in high-dimensional environments and unknown system dynamics, i.e., model-free RL, the solution of parametric density estimation is only tractable up to the density normalization constant. To solve this, we formulate the problem as a mixture of logistic regressions to directly handle the unnormalized density. To research the challenges faced by overlapping demonstrations, we introduce the concepts of shared pair, which is a state-action pair that is shared in more than one intention, and separability, which resembles how well the multiple intentions can be separated in the joint state-action space. We provide theoretical analyses under the global optimality condition and the existence of shared pairs. Furthermore, we conduct extensive experiments on four simulated robotics tasks, extended to accept different intentions with specific levels of separability, and a synthetic driver task developed to directly control the separability. We evaluate the existing baselines on our defined problem and demonstrate, theoretically and experimentally, the advantages of our clustering-based solution, especially when the separability of the demonstrations decreases.

We address the problem of inverse reinforcement learning in Markov decision processes where the agent is risksensitive. In particular, we model risk-sensitivity in a reinforcement learning framework by making use of models of human decision-making having their origins in behavioral psychology and economics. We propose a gradient-based inverse reinforcement learning algorithm that minimizes a loss function defined on the observed behavior. We demonstrate the performance of the proposed technique on two examples, the first of which is the canonical Grid World example and the second of which is an MDP modeling passengers' decisions regarding ride-sharing. In the latter, we use pricing and travel time data from a ride-sharing company to construct the transition probabilities and rewards of the MDP.

Learning from demonstration, or imitation learning, is the process of learning to act in an environment from examples provided by a teacher. Inverse reinforcement learning (IRL) is a specific form of learning from demonstration that attempts to estimate the reward function of a Markov decision process from examples provided by the teacher. The reward function is often considered the most succinct description of a task. In simple applications, the reward function may be known or easily derived from properties of the system and hard coded into the learning process. However, in complex applications, this may not be possible, and it may be easier to learn the reward function by observing the actions of the teacher. This paper provides a comprehensive survey of the literature on IRL. This survey outlines the differences between IRL and two similar methods - apprenticeship learning and inverse optimal control. Further, this survey organizes the IRL literature based on the principal method, describes applications of IRL algorithms, and provides areas of future research.

Introduction Reinforcement learning formalizes the concept of learning from interactions.1 Broadly, reinforcement learning focuses on a setting in which an agent (decision maker) sequentially interacts with an environment that is partially unknown to them. At each stage, the agent takes an action and receives a reward. The objective of the agent is to maximize rewards accumulated in the long run. There are many situations in health care where decisions are made sequentially for which reinforcement learning approaches could prove useful for decision making. Throughout this article, we consider treatment prescription as an archetypical example to connect reinforcement learning concepts to a health care setting. In this setting, the care provider, the prescribed treatment, and the patients can be viewed as the agent, the action, and the environment, respectively, as depicted in Figure 1.Figure 1: Sequential treatment of AKI or CKD complications modeled as a reinforcement learning problem.Background In this section, with the objective of making reinforcement learning literature more accessible to a clinical audience, we briefly introduce related fundamental concepts and approaches. We refer the interested reader to Sutton and Barto1 for a comprehensive introduction to reinforcement learning. Markov Decision Processes Markov decision processes (MDPs) are a formalism of the sequential decision-making problem that has been central to the theoretical and practical advancements of reinforcement learning. In each stage of an MDP, the agent observes the state of the environment and takes an action, which, in turn, results in a change of the state. This change of state is assumed to be probabilistic with the next state being determined only by the preceding state, the chosen action, and the transition probability. The agent also receives a reward that is a function of the taken action, the preceding state, and the subsequent state. In an MDP, the objective of the agent is to maximize the return defined as the reward accumulated over a time horizon. In some applications, it is common to consider the horizon to be infinite, in which case the future rewards are discounted by a factor smaller than one. The selection of action by the agent on the basis of the observed state is known as the policy. More formally, a policy is a probabilistic mapping from states to each possible action. Because the policy and the reward are a function of the state, it is critical to estimate the utility of being in a certain state. More specifically, the value function is defined as the expected return starting from a given state under the chosen policy. Under this formalism, the objective of the agent is to find the optimal policy that maximizes the value function for all states. Reinforcement Learning Methods Action-value methods are a class of reinforcement learning methods in which the actions are chosen on the basis of the estimation of their long-term value. A prominent example of an action-value method is Q-learning in which the agent iteratively takes actions with the highest estimated values and updates the action-state value function on the basis of new observations. Policy gradient methods are another class of reinforcement learning methods that seek to optimize the policy directly instead of choosing actions on the basis of their respective estimated value. Such methods could be advantageous in health care applications that entail a large number of possible actions, e.g., when recommending a wide range of drug dosages or treatment options. Clinical Applications Reinforcement learning frameworks and methods are broadly applicable to clinical settings in which decisions are made sequentially. A prominent clinical application of reinforcement learning is for treatment recommendation, which has been studied across a variety of diseases and treatments including radiation and chemotherapy for cancer, brain stimulation for epilepsy, and treatment strategies for sepsis.2–5 In such treatment recommendation settings, a policy is commonly known as a dynamic treatment regime. There are various other clinical applications of reinforcement learning including diagnosis, medical imaging, and decision support tools (see refs. 2–5 and the references therein). Reinforcement Learning in Nephrology Although there have been recent applications of machine learning in nephrology,6,7 to the best of the authors' knowledge, the application of reinforcement learning to nephrology has been primarily limited to optimizing the erythropoietin dosage in hemodialysis patients.8,9 However, there are other settings where reinforcement learning has the potential to improve patient care in nephrology. For example, reinforcement learning methods can be adopted in the treatment of the complications of AKI or CKD (Figure 1). In this problem, the state models the conditions of the patient (e.g., vital signs, laboratory test results including urine and blood tests, and urine output measurements). The action refers to the treatment options (e.g., the dosage of medications such as sodium polystyrene sulfonate, and hemodialysis). The reward models the improvement in patient conditions. Similarly, reinforcement learning can help automate and optimize the dosage of immunosuppressive drugs in kidney transplants. Challenges and Opportunities Despite the success of reinforcement learning in several simplified clinical settings, their large-scale application to patient care faces several open challenges. The complexity of human biology complicates modeling clinical decision making as a reinforcement learning problem. The state space in such settings is often enormous, which could make a purely computational approach infeasible. Moreover, modeling all potential objectives a priori as a reward function may not be feasible. To overcome these challenges and realize the potential of reinforcement learning, clinical insight can play a pivotal role. More specifically, restricting the state space to only include highly relevant clinical variables could greatly reduce the computational complexity. Furthermore, using inverse reinforcement learning,2 relevant reward functions can be learned from retrospective studies assuming the optimality of clinical decisions. Another critical challenge is addressing moral and ethical concerns. It is imperative to ensure that reinforcement learning methods do not cause harm to the patient. To this end, there exists a need for a thorough validation of such methods before their use in patient care. Hence, there is a need to go beyond retrospective studies that have been used for the proof of concept of most existing reinforcement learning methods in health care applications.2,3 The lessons learned from the success of reinforcement learning in other application areas (e.g., self-driving cars) can help navigate the path to realizing its potential in health care. Accessible open-source simulation environments that enable researchers to compare various approaches are essential to the field of reinforcement learning. OpenAI Gym is currently the leading toolkit containing a wide range of simulated environments, e.g., surgical robotics.10 The development of high-quality and reliable simulation environments for nephrology and other health care applications can facilitate the development and validation of reinforcement learning methods beyond limited retrospective studies. The adoption of methods validated in such simulation environments in actual clinical settings will require clinicians' oversight. Similar to how self-driving cars require a human driver to ensure collision avoidance, clinicians' oversight is critical to ensure the safety of the patients, especially in the early stages of the adoption of reinforcement learning methods. The data from clinicians' decisions (e.g., overruling the automated treatment recommendation) can be used to improve the reliability of autonomous systems over time and reduce the burden of clinicians' oversight.

In this work, we proposed a novel Hybrid Reinforcement Learning algorithm that incorporates the model-based reinforcement learning scheme with the model-free reinforcement learning framework, where the sample efficiency was improved while still retaining the desired control performance. The proposed learning framework was successfully applied to a complex biped robot model, where the robot was able to maintain balance on a rotating platform. Experiment results also showed that the proposed algorithm was able to generate stable walking gaits for the robot to walk on both static and dynamic platforms without falling down.

Reinforcement learning (RL) provides a computational model to animal’s autonomous acquisition of behaviors even in an uncertain environment. Inverse reinforcement learning (IRL) is its opposite; given a history of behaviors of an agent, IRL attempts to determine the unknown characteristics, like a reward function, of the agent. Conventional IRL methods usually assume the agent has taken a stationary policy that is optimal in the environment. However, real RL agents do not necessarily take stationary policy, because they are often on the way of adapting to their own environments. Especially when facing an uncertain environment, an intelligent agent should take a mixed (or switching) strategy consisting of an exploitation that is best at the current situation and an exploration to resolve the environmental uncertainty. In this study, we propose a new IRL method that can identify both of a non-stationary policy and a fixed but unknown reward function, based on the behavioral history of a learning agent; in particular, we estimate a change point of the behavior policy from an exploratory one in the agent’s early stage of the learning and an exploitative one in its later learning stage. When applied to a computer simulation during a simple maze task of an agent, our method could identify the change point of the behavior policy and the fixed reward function, only from the agent’s history of behaviors.

To behave adaptively, we must learn from the consequences of our actions. Doing so is difficult when the consequences of an action follow a delay. This introduces the problem of temporal credit assignment. When feedback follows a sequence of decisions, how should the individual assign credit to the intermediate actions that comprise the sequence? Research in reinforcement learning provides 2 general solutions to this problem: model-free reinforcement learning and model-based reinforcement learning. In this review, we examine connections between stimulus-response and cognitive learning theories, habitual and goal-directed control, and model-free and model-based reinforcement learning. We then consider a range of problems related to temporal credit assignment. These include second-order conditioning and secondary reinforcers, latent learning and detour behavior, partially observable Markov decision processes, actions with distributed outcomes, and hierarchical learning. We ask whether humans and animals, when faced with these problems, behave in a manner consistent with reinforcement learning techniques. Throughout, we seek to identify neural substrates of model-free and model-based reinforcement learning. The former class of techniques is understood in terms of the neurotransmitter dopamine and its effects in the basal ganglia. The latter is understood in terms of a distributed network of regions including the prefrontal cortex, medial temporal lobes, cerebellum, and basal ganglia. Not only do reinforcement learning techniques have a natural interpretation in terms of human and animal behavior but they also provide a useful framework for understanding neural reward valuation and action selection.

Advanced planning for autonomous vehicles using reinforcement learning and deep inverse reinforcement learning

Reinforcement learning, characterized by trial-and-error learning and delayed rewards, is central to decision-making processes. Its core component, the reward function, is traditionally handcrafted, but designing these functions is often challenging or impossible in real-world scenarios. Inverse reinforcement learning (IRL) addresses this issue by extracting reward functions from expert demonstrations, facilitating optimal policy derivation and offering a deeper understanding of expert behavior. This comprehensive review focuses on three key aspects: the diverse methodologies employed in IRL, its wide-ranging applications across fields such as robotics, autonomous vehicles, and human intent analysis, and the importance of curated datasets in advancing IRL research. A structured analysis of IRL techniques is provided, applications are categorized by domain, and the role of benchmark datasets in evaluating performance and guiding future developments is emphasized. The unique value of IRL in bridging the gap between human and artificial learning is highlighted, demonstrating its potential to unlock advancements in machine learning, decision making, and explainable AI. By summarizing the current state of IRL research and advocating for future directions, this review serves as a valuable resource for researchers and practitioners seeking to explore and advance the field.

Reinforcement Learning (RL) Markov Decision Processes is studied with an emphasis on the well-studied exploration problem. We first formulate and discuss a definition of “efficient” algorithms that is termed Probably Approximately Correct (PAC) in RL. Next we provide general sufficient conditions for such an algorithm that applies to several different modeling assumptions. The conditions can be used to demonstrate that efficient learning is possible in finite MDPs, with either a model-based or model-free approach, in factored MDPs, and in continuous MDPs with linear dynamics. In the reinforcement-learning (RL) problem (Sutton & Barto 1998), an agent acts in an unknown or incompletely known environment with the goal of maximizing an external reward signal. In the most standard mathematical formulation of the problem, the environment is modeled as a Markov Decision Process (MDP) and the goal of the agent is to obtain near-optimal discounted return. Over the years, many algorithms have been proposed for this problem, but analyses of their performances have been relatively scarce. In fact, until recently, most theoretical guarantees have been that certain algorithms will discover an optimal policy in the limit, after an infinite amount of experience. In contrast, several attempts have been made to study “Probably Approximately Correct” or PAC-MDP algorithms, which exhibit near-optimal behavior in polynomial time and experience. This paper discusses several extensions of those results. We present a theorem that provides sufficient conditions for an algorithm to be PAC-MDP. We examine these conditions and show how they can be applied to prove that efficient learning is possible in three interesting scenerios: finite MDPs (i.e. the “Tabular case”),