Imitation learning (IL) frameworks in robotics typically assume that a domain expert's demonstration always contains a correct way of doing the task. Despite its theoretical convenience, this assumption has limited practical values for an IL-powered robot in real world. There are many reasons for an expert in the real world to provide demonstrations that may contain incorrect or potentially unsafe way of doing a task. In order for IL-powered robots to work in the real world, IL frameworks need to detect such adversarial demonstrations and not learn from them. This paper proposes an IL framework that can autonomously detect and remove adversarial demonstrations, if they exist in the demonstration set, as it directly learns a task policy from the expert. The proposed framework that we term Robust Maximum Entropy behavior cloning (R-MaxEnt) learns a stochastic model that maps states to actions. In doing so, R-MaxEnt solves a minmax problem that leverages the entropy of the model to assign weights to different demonstrations while assigning poor weights to adversarial samples. Our empirical results show that R-MaxEnt outperforms the existing IL approaches in both real and simulated robotics tasks.

High-level human activities often have rich temporal structures that determine the order in which atomic actions are executed. We propose the Temporal Context Graph (TCG), a temporal reasoning model that integrates probabilistic inference with Allen's interval algebra, to capture these temporal structures. TCGs are capable of modeling tasks with cyclical atomic actions and consisting of sequential and parallel temporal relations. We present Learning from Demonstration as the application domain where the use of TCGs can improve policy selection and address the problem of perceptual aliasing. Experiments validating the model are presented for learning two tasks from demonstration that involve structured human-robot interactions. The source code for this implementation is available at https://github.com/AssistiveRoboticsUNH/TCG.

There are many human-robot interaction (HRI) tasks that are highly structured and follow a certain temporal sequence. Learning such tasks from demonstrations requires understanding the underlying rules governing the interactions. This involves identifying and generalizing the key spatial and temporal features of the task and capturing the high-level relationships among them. Despite its crucial role in sequential task learning, temporal reasoning is often ignored in existing learning from demonstration (LFD) research. This paper proposes a holistic LFD framework that learns the underlying temporal structure of sequential HRI tasks. The proposed Temporal-Reasoning-based LFD (TR-LFD) framework relies on an automated spatial reasoning layer to identify and generalize relevant spatial features, and a temporal reasoning layer to analyze and learn the high-level temporal structure of a HRI task. We evaluate the performance of this framework by learning a well-explored task in HRI research: robot-mediated autism intervention. The source code for this implementation is available at https://github.com/AssistiveRoboticsUNH/TR-LFD.

Extracting a set of generalizable rules that govern the dynamics of complex, high-level interactions between humans based only on observations is a high-level cognitive ability. Mastery of this skill marks a significant milestone in the human developmental process. A key challenge in designing such an ability in autonomous robots is discovering the relationships among discriminatory features. Identifying features in natural scenes that are representative of a particular event or interaction (i.e. »discriminatory features») and then discovering the relationships (e.g., temporal/spatial/spatio-temporal/causal) among those features in the form of generalized rules are non-trivial problems. They often appear as a »chicken-and-egg» dilemma. This paper proposes an end-to-end learning framework to tackle these two problems in the context of learning generalized, high-level rules of human interactions from structured demonstrations. We employed our proposed deep reinforcement learning framework to learn a set of rules that govern a behavioral intervention session between two agents based on observations of several instances of the session. We also tested the accuracy of our framework with human subjects in diverse situations.

The decentralized partially observable Markov decision process (Dec-POMDP) is a powerful model for representing multi-agent problems with decentralized behavior. Unfortunately, current Dec-POMDP solution methods cannot solve problems with continuous observations, which are common in many real-world domains. To that end, we present a framework for representing and generating Dec-POMDP policies that explicitly include continuous observations. We apply our algorithm to a novel tagging problem and an extended version of a common benchmark, where it generates policies that meet or exceed the values of equivalent discretized domains without the need for finding an adequate discretization.

We present a learning from demonstration (LfD) framework that uses a deep recurrent Q-network (DRQN) to learn how to deliver a behavioral intervention (BI) from demonstrations performed by a human. The trained DRQN enables a robot to deliver a similar BI in an autonomous manner. BIs are highly structured procedures wherein children with developmental delays/disorders (e.g. autism, ADHD, etc.) are trained to perform new behaviors and life-skills. Mounting anecdotal evidence from human-robot interaction (HRI) research has shown that BI benefits from the use of robots as a delivery tool. Most of the HRI research on robot-based intervention relies on tele-operated robots. However, the need for autonomy has become increasingly evident, especially when it comes to the real-world deployment of these systems. The few studies that have used autonomy in robot-based BI relied on hand-picked features of the environment in order to trigger correct robot actions. Additionally, none of these automated architectures attempted to learn the BI from human demonstrations, though this appears to be the most natural way of learning. This paper represents the first attempt to design a robot that uses LfD to learn BI. We generate a model then correctly predict appropriate actions with greater than 80% accuracy. To the best of our knowledge, this is the first attempt to employ DRQN within an LfD framework to learn high level reasoning embedded in human actions and behaviors simply from observations.