We present a comprehensive survey of robot Learning from Demonstration (LfD), a technique that develops policies from example state to action mappings. We introduce the LfD design choices in terms of demonstrator, problem space, policy derivation and performance, and contribute the foundations for a structure in which to categorize LfD research. Specifically, we analyze and categorize the multiple ways in which examples are gathered, ranging from teleoperation to imitation, as well as the various techniques for policy derivation, including matching functions, dynamics models and plans. To conclude we discuss LfD limitations and related promising areas for future research.

We present Confidence-Based Autonomy (CBA), an interactive algorithm for policy learning from demonstration. The CBA algorithm consists of two components which take advantage of the complementary abilities of humans and computer agents. The first component, Confident Execution, enables the agent to identify states in which demonstration is required, to request a demonstration from the human teacher and to learn a policy based on the acquired data. The algorithm selects demonstrations based on a measure of action selection confidence, and our results show that using Confident Execution the agent requires fewer demonstrations to learn the policy than when demonstrations are selected by a human teacher. The second algorithmic component, Corrective Demonstration, enables the teacher to correct any mistakes made by the agent through additional demonstrations in order to improve the policy and future task performance. CBA and its individual components are compared and evaluated in a complex simulated driving domain. The complete CBA algorithm results in the best overall learning performance, successfully reproducing the behavior of the teacher while balancing the tradeoff between number of demonstrations and number of incorrect actions during learning. 1.

We contribute Policy Reuse as a technique to improve a reinforcement learning agent with guidance from past learned similar policies. Our method relies on using the past policies as a probabilistic bias where the learning agent faces three choices: the exploitation of the ongoing learned policy, the exploration of random unexplored actions, and the exploitation of past policies. We introduce the algorithm and its major components: an exploration strategy to include the new reuse bias, and a similarity function to estimate the similarity of past policies with respect to a new one. We provide empirical results demonstrating that Policy Reuse improves the learning performance over different strategies that learn without reuse. Interestingly and almost as a side effect, Policy Reuse also identifies classes of similar policies revealing a basis of core policies of the domain. We demonstrate that such a basis can be built incrementally, contributing the learning of the structure of a domain.

Abstract — We tackle the problem of learning imitative wholebody motions in a humanoid robot using probabilistic inference in Bayesian networks. Our inference-based approach affords a straightforward method to exploit rich yet uncertain prior information obtained from human motion capture data. Dynamic imitation implies that the robot must interact with its environment and account for forces such as gravity and inertia during imitation. Rather than explicitly modeling these forces and the body of the humanoid as in traditional approaches, we show that stable imitative motion can be achieved by learning a sensorbased representation of dynamic balance. Bayesian networks provide a sound theoretical framework for combining prior kinematic information (from observing a human demonstrator) with prior dynamic information (based on previous experience) to model and subsequently infer motions which, with high probability, will be dynamically stable. By posing the problem as one of inference in a Bayesian network, we show that methods developed for approximate inference can be leveraged to efficiently perform inference of actions. Additionally, by using nonparametric inference and a nonparametric (Gaussian process) forward model, our approach does not make any strong assumptions about the physical environment or the mass and inertial properties of the humanoid robot. We propose an iterative, probabilistically constrained algorithm for exploring the space of motor commands and show that the algorithm can quickly discover dynamically stable actions for whole-body imitation of human motion. Experimental results based on simulation and subsequent execution by a HOAP-2 humanoid robot demonstrate that our algorithm is able to imitate a human performing actions such as squatting and a one-legged balance. I.

As learning agents move from research labs to the real world, it is increasingly important that human users, including those without programming skills, be able to teach agents desired behaviors. Recently, the tamer framework was introduced for designing agents that can be interactively shaped by human trainers who give only positive and negative feedback signals. Past work on tamer showed that shaping can greatly reduce the sample complexity required to learn a good policy, can enable lay users to teach agents the behaviors they desire, and can allow agents to learn within a Markov Decision Process (MDP) in the absence of a coded reward function. However, tamer does not allow this human training to be combined with autonomous learning based on such a coded reward function. This paper

Abstract not found

An ambitious goal of transfer learning is to learn a task faster after training on a different, but related, task. In this paper we extend a previously successful temporal difference (Sutton & Barto, 1998) approach to transfer in reinforcement learning (Sutton & Barto, 1998) tasks to work with policy search. In particular, we show how to construct a mapping to translate a population of policies trained via genetic algorithms (GAs) (Goldberg, 1989) from a source task to a target task. Empirical results in robot soccer Keepaway, a standard RL benchmark domain (Stone et al., 2006), demonstrate that transfer via inter-task mapping can markedly reduce the time required to learn a second, more complex, task.

In this paper, we combine the formal methods from reinforcement learning with the paradigm of imitation learning. The extension of the reinforcement learning framework to integrate the information provided by an expert (demonstrator) has the important advantage of allowing a clear decrease of the time necessary to learn certain robotic tasks. Hence, learning by imitation can be interpreted as a mechanism for fast skill transfer. Another contribution of thispaper consists in showing that our formalism is able to model different types of imitation-learning that are described in the biological literature. It thus unifies in the same abstract model what used to be addressed as separate behavioral patterns. We illustrate the application of these methods in simulation and with a real robot.

This chapter reviews several approaches to the problem of learning by imitation in robotics. We start by describing several cognitive processes identified in the literature as necessary for imitation. We then proceed by surveying different approaches to this problem, placing particular emphasys on methods whereby an agent first learns about its own body dynamics by means of self-exploration and then uses this knowledge about its own body to recognize the actions being performed by other agents. This general approach is related to the motor theory of perception, particularly to the mirror neurons found in primates. We distinguish three fundamental classes of methods, corresponding to three abstraction levels at which imitation can be addressed. As such, the methods surveyed herein exhibit behaviors that range from raw sensory-motor trajectory matching to high-level abstract task replication. We also discuss the impact that knowledge about the world and/or the demonstrator can have on the particular behaviors exhibited.

Abstract. Imitation-based learning is a general mechanism for rapid acquisition of new behaviors in autonomous agents and robots. In this paper, we propose a new approach to learning by imitation based on parameter learning in probabilistic graphical models. Graphical models are used not only to model an agent’s own dynamics but also the dynamics of an observed teacher. Parameter tying between the agent-teacher models ensures consistency and facilitates learning. Given only observations of the teacher’s states, we use the expectation-maximization (EM) algorithm to learn both dynamics and policies within graphical models. We present results demonstrating that EM-based imitation learning outperforms pure exploration-based learning on a benchmark problem (the FlagWorld domain). We additionally show that the graphical model representation can be leveraged to incorporate domain knowledge (e.g., state space factoring) to achieve significant speed-up in learning. 1