This paper introduces BarlowRL, a data-efficient reinforcement learning agent that combines the Barlow Twins self-supervised learning framework with DER (Data-Efficient Rainbow) algorithm. BarlowRL outperforms both DER and its contrastive counterpart CURL on the Atari 100k benchmark. BarlowRL avoids dimensional collapse by enforcing information spread to the whole space. This helps RL algorithms to utilize uniformly spread state representation that eventually results in a remarkable performance. The integration of Barlow Twins with DER enhances data efficiency and achieves superior performance in the RL tasks. BarlowRL demonstrates the potential of incorporating self-supervised learning techniques, especially that of non-contrastive objectives, to improve RL algorithms.

This paper introduces a novel Learning from Demonstration framework to learn robotic skills with keyframe demonstrations using a Dynamic Bayesian Network (DBN) and a Bayesian Optimized Policy Search approach to improve the learned skills. DBN learns the robot motion, perceptual change in the object of interest (aka skill sub-goals) and the relation between them. The rewards are also learned from the perceptual part of the DBN. The policy search part is a semiblack box algorithm, which we call BO-PI2 . It utilizes the action-perception relation to focus the high-level exploration, uses Gaussian Processes to model the expected-return and performs Upper Confidence Bound type low-level exploration for sampling the rollouts. BO-PI2 is compared against a stateof-the-art method on three different skills in a real robot setting with expert and naive user demonstrations. The results show that our approach successfully focuses the exploration on the failed sub-goals and the addition of reward-predictive exploration outperforms the state-of-the-art approach on cumulative reward, skill success, and termination time metrics.

Traditional approaches to extrinsic calibration use fiducial markers and learning-based approaches rely heavily on simulation data. In this work, we present a learning-based markerless extrinsic calibration system that uses a depth camera and does not rely on simulation data. We learn models for end-effector (EE) segmentation, single-frame rotation prediction and keypoint detection, from automatically generated real-world data. We use a transformation trick to get EE pose estimates from rotation predictions and a matching algorithm to get EE pose estimates from keypoint predictions. We further utilize the iterative closest point algorithm, multiple-frames, filtering and outlier detection to increase calibration robustness. Our evaluations with training data from multiple camera poses and test data from previously unseen poses give sub-centimeter and sub-deciradian average calibration and pose estimation errors. We also show that a carefully selected single training pose gives comparable results.

The goal of this research is to enable robots to learn new things from everyday people. For years, the AI and Robotics community has sought to enable robots to efficiently learn new skills from a knowledgeable human trainer, and prior work has focused on several important technical problems. This vast amount of research in the field of robot Learning by Demonstration has by and large only been evaluated with expert humans, typically the system's designer. Thus, neglecting a key point that this interaction takes place within a social structure that can guide and constrain the learning problem. %Moreover, we We believe that addressing this point will be essential for developing systems that can learn from everyday people that are not experts in Machine Learning or Robotics. Our work focuses on new research questions involved in letting robots learn from everyday human partners (e.g., What kind of input do people want to provide a machine learner? How does their mental model of the learning process affect this input? What interfaces and interaction mechanisms can help people provide better input from a machine learning perspective?) Often our research begins with an investigation into the feasibility of a particular machine learning interaction, which leads to a series of research questions around re-designing both the interaction and the algorithm to better suit learning with end-users. We believe this equal focus on both the Machine Learning and the HRI contributions are key to making progress toward the goal of machines learning from humans. In this abstract we briefly overview four different projects that highlight our HRI approach to the problem of Learning from Demonstration.

The field of robot Learning from Demonstration (LfD) makes use of several input modalities for demonstrations (teleoperation, kinesthetic teaching, marker- and vision-based motion tracking). In this paper we present two experiments aimed at identifying and overcoming challenges associated with using teleoperation as an input modality for LfD. Our first experiment compares kinesthetic teaching and teleoperation and highlights some inherent problems associated with teleoperation; specifically uncomfortable user interactions and inaccurate robot demonstrations. Our second experiment is focused on overcoming these problems and designing the teleoperation interaction to be more suitable for LfD. In previous work we have proposed a novel demonstration strategy using the concept of keyframes, where demonstrations are in the form of a discrete set of robot configurations. Keyframes can be naturally combined with continuous trajectory demonstrations to generate a hybrid strategy. We perform user studies to evaluate each of these demonstration strategies individually and show that keyframes are intuitive to the users and are particularly useful in providing noise-free demonstrations. We find that users prefer the hybrid strategy best for demonstrating tasks to a robot by teleoperation.

Robot Learning from Demonstration (LfD) research deals with the challenges of enabling humans to teach robots novel skills and tasks (Argall et al. 2009). The practical importance of LfD is due to the fact that it is impossible to pre-program all the necessary skills and task knowledge that a robot might need during its life-cycle. This poses many interesting application areas for LfD ranging from houses to factory floors. An important motivation for our research agenda is that in many of the practical LfD applications, the teacher will be an everyday end-user, not an expert in Machine Learning or robotics. Thus, our research explores the ways in which Machine Learning can exploit human social learning interactions--Socially Guided Machine Learning (SGML).