Interactive Machine Learning

When faced with accomplishing a task, human experts exhibit intentional behavior. Their unique intents shape their plans and decisions, resulting in experts demonstrating diverse behaviors to accomplish the same task. Due to the uncertainties encountered in the real world and their bounded rationality, experts sometimes adjust their intents, which in turn influences their behaviors during task execution. This paper introduces IDIL, a novel imitation learning algorithm to mimic these diverse intent-driven behaviors of experts. Iteratively, our approach estimates expert intent from heterogeneous demonstrations and then uses it to learn an intent-aware model of their behavior. Unlike contemporary approaches, IDIL is capable of addressing sequential tasks with high-dimensional state representations, while sidestepping the complexities and drawbacks associated with adversarial training (a mainstay of related techniques). Our empirical results suggest that the models generated by IDIL either match or surpass those produced by recent imitation learning benchmarks in metrics of task performance. Moreover, as it creates a generative model, IDIL demonstrates superior performance in intent inference metrics, crucial for human-agent interactions, and aptly captures a broad spectrum of expert behaviors.

Offline imitation learning (IL) refers to learning expert behavior solely from demonstrations, without any additional interaction with the environment. Despite significant advances in offline IL, existing techniques find it challenging to learn policies for long-horizon tasks and require significant re-training when task specifications change. Towards addressing these limitations, we present GO-DICE an offline IL technique for goal-conditioned long-horizon sequential tasks. GO-DICE discerns a hierarchy of sub-tasks from demonstrations and uses these to learn separate policies for sub-task transitions and action execution, respectively; this hierarchical policy learning facilitates long-horizon reasoning. Inspired by the expansive DICE-family of techniques, policy learning at both the levels transpires within the space of stationary distributions. Further, both policies are learnt with goal conditioning to minimize need for retraining when task goals change. Experimental results substantiate that GO-DICE outperforms recent baselines, as evidenced by a marked improvement in the completion rate of increasingly challenging pick-and-place Mujoco robotic tasks. GO-DICE is also capable of leveraging imperfect demonstration and partial task segmentation when available, both of which boost task performance relative to learning from expert demonstrations alone.

We present Bayesian Team Imitation Learner (BTIL), an imitation learning algorithm to model the behavior of teams performing sequential tasks in Markovian domains. In contrast to existing multi-agent imitation learning techniques, BTIL explicitly models and infers the time-varying mental states of team members, thereby enabling learning of decentralized team policies from demonstrations of suboptimal teamwork. Further, to allow for sample- and label-efficient policy learning from small datasets, BTIL employs a Bayesian perspective and is capable of learning from semi-supervised demonstrations. We demonstrate and benchmark the performance of BTIL on synthetic multi-agent tasks as well as a novel dataset of human-agent teamwork. Our experiments show that BTIL can successfully learn team policies from demonstrations despite the influence of team members’ (time-varying and potentially misaligned) mental states on their behavior.

Autonomous agents operating in the real world often need to interact with other agents to accomplish their tasks. For such agents, the ability to model behavior of other agents – both human and artificial – without complete knowledge of their decision factors is essential. Towards realizing this ability, we present Factorial Agent Markov Model (FAMM), a model to represent behavior of other agents performing sequential tasks. In contrast with most existing models, FAMM allows for behavior of other agents to depend on multiple, time-varying latent decision factors and does not assume rationality. To enable learning of \famm parameters by observing behavior of other agents, we provide a set of variational inference algorithms for the unsupervised, semi-supervised, and supervised settings. These Bayesian learning algorithms for the FAMM enable agents to model other agents using execution traces and domain-specific priors. We demonstrate the utility of FAMM and corresponding learning algorithms using three synthetic domains and benchmark them against existing algorithms for modeling agent behavior. Our numerical experiments demonstrate that, despite the presence of multiple and time-varying latent states, our approach is capable of learning predictive models of other agents with semi-supervision.

Rewards play a crucial role in reinforcement learning. To arrive at the desired policy, the design of a suitable reward function often requires significant domain expertise as well as trial-and-error. Here, we aim to minimize the effort involved in designing reward functions for contact-rich manipulation tasks. In particular, we provide an approach capable of extracting dense reward functions algorithmically from robots’ high-dimensional observations, such as images and tactile feedback. In contrast to state-of-the-art high-dimensional reward learning methodologies, our approach does not leverage adversarial training, and is thus less prone to the associated training instabilities. Instead, our approach learns rewards by estimating task progress in a self-supervised manner. We demonstrate the effectiveness and efficiency of our approach on two contact-rich manipulation tasks, namely, peg-in-hole and USB insertion. The experimental results indicate that the policies trained with the learned reward function achieves better performance and faster convergence compared to the baselines.

Artificial agents that interact with other (human or artificial) agents require models in order to reason about those other agents’ behavior. In addition to the predictive utility of these models, maintaining a model that is aligned with an agent’s true generative model of behavior is critical for effective human-agent interaction. In applications wherein observations and partial specification of the agent’s behavior are available, achieving model alignment is challenging for a variety of reasons. For one, the agent’s decision factors are often not completely known; further, prior approaches that rely upon observations of agents’ behavior alone can fail to recover the true model, since multiple models can explain observed behavior equally well. To achieve better model alignment, we provide a novel approach capable of learning aligned models that conform to partial knowledge of the agent’s behavior. Central to our approach are a factored model of behavior (AMM), along with Bayesian nonparametric priors, and an inference approach capable of incorporating partial specifications as constraints for model learning. We evaluate our approach in experiments and demonstrate improvements in metrics of model alignment.