This paper presents an end-to-end self-supervised learning approach for cross-modality image registration and homography estimation, with a particular emphasis on registering sports field templates onto broadcast videos as a practical application. Rather then using any pairwise labelled data for training, we propose a self-supervised data mining method to train the registration network with a natural image and its edge map. Using an iterative estimation process controlled by a score regression network (SRN) to measure the registration error, the network can learn to estimate any homography transformation regardless of how misaligned the image and the template is. We further show the benefits of using pretrained weights to finetune the network for sports field calibration with few training data. We demonstrate the effectiveness of our proposed method by applying it to real-world sports broadcast videos where we achieve state-of-the-art results and real-time processing.

We consider the task of underwater robot navigation for the purpose of collecting scientifically relevant video data for environmental monitoring. The majority of field robots that currently perform monitoring tasks in unstructured natural environments navigate via path-tracking a pre-specified sequence of waypoints. Although this navigation method is often necessary, it is limiting because the robot does not have a model of what the scientist deems to be relevant visual observations. Thus, the robot can neither visually search for particular types of objects, nor focus its attention on parts of the scene that might be more relevant than the pre-specified waypoints and viewpoints. In this paper we propose a method that enables informed visual navigation via a learned visual similarity operator that guides the robot's visual search towards parts of the scene that look like an exemplar image, which is given by the user as a high-level specification for data collection. We propose and evaluate a weakly supervised video representation learning method that outperforms ImageNet embeddings for similarity tasks in the underwater domain. We also demonstrate the deployment of this similarity operator during informed visual navigation in collaborative environmental monitoring scenarios, in large-scale field trials, where the robot and a human scientist collaboratively search for relevant visual content. Code: https://github.com/rvl-lab-utoronto/visual_search_in_the_wild.

This paper considers the problem of learning behaviors in simulation without knowledge of the precise dynamical properties of the target robot platform(s). In this context, our learning goal is to mutually maximize task efficacy on each environment considered and generalization across the widest possible range of environmental conditions. The physical parameters of the simulator are modified by a component of our technique that learns the Domain Randomization (DR) that is appropriate at each learning epoch to maximally challenge the current behavior policy, without being overly challenging, which can hinder learning progress. This so-called sweet spot distribution is a selection of simulated domains with the following properties: 1) The trained policy should be successful in environments sampled from the domain randomization distribution; and 2) The DR distribution made as wide as possible, to increase variability in the environments. These properties aim to ensure the trajectories encountered in the target system are close to those observed during training, as existing methods in machine learning are better suited for interpolation than extrapolation. We show how adapting the DR distribution while training context-conditioned policies results in improvements on jump-start and asymptotic performance when transferring a learned policy to the target environment1.

We propose an optimization-based framework to register sports field templates onto broadcast videos. For accurate registration we go beyond the prevalent feed-forward paradigm. Instead, we propose to train a deep network that regresses the registration error, and then register images by finding the registration parameters that minimize the regressed error. We demonstrate the effectiveness of our method by applying it to real-world sports broadcast videos, outperforming the state of the art. We further apply our method on a synthetic toy example and demonstrate that our method brings significant gains even when the problem is simplified and unlimited training data is available1.

Diversity of environments is a key challenge that causes learned robotic controllers to fail due to the discrepancies between the training and evaluation conditions. Training from demonstrations in various conditions can mitigate - but not completely prevent - such failures. Learned controllers such as neural networks typically do not have a notion of uncertainty that allows to diagnose an offset between training and testing conditions, and potentially intervene. In this work, we propose to use Bayesian Neural Networks, which have such a notion of uncertainty. We show that uncertainty can be leveraged to consistently detect situations in high-dimensional simulated and real robotic domains in which the performance of the learned controller would be sub-par. Also, we show that such an uncertainty based solution allows making an informed decision about when to invoke a fallback strategy. One fallback strategy is to request more data. We empirically show that providing data only when requested results in increased data-efficiency.

We present an algorithm for rapidly learning neural network policies for robotics systems. The algorithm follows the model-based reinforcement learning paradigm and improves upon existing algorithms: PILeO and a sample-based version of PILeo with neural network dynamics (Deep-PILeO). To improve convergence, we propose a model-based algorithm that uses fixed random numbers and clips gradients during optimization. We propose training a neural network dynamics model using variational dropout with truncated Log-Normal noise. These improvements enable data-efficient synthesis of complex neural network policies. We test our approach on a variety of benchmark tasks, demonstrating data-efficiency that is competitive with that of PILeO, while being able to optimize complex neural network controllers. Finally, we assess the performance of the algorithm for learning motor controllers for a six legged autonomous underwater vehicle. This demonstrates the potential of the algorithm for scaling up the dimensionality and dataset sizes, in more complex tasks.

We address the problem of learning vision-based, collision-avoiding, and target-selecting controllers in 3D, specifically in underwater environments densely populated with coral reefs. Using a highly maneuverable, dynamic, six-legged (or flippered) vehicle to swim underwater, we exploit real time visual feedback to make close-range navigation decisions that would be hard to achieve with other sensors. Our approach uses computer vision as the sole mechanism for both collision avoidance and visual target selection. In particular, we seek to swim close to the reef to make observations while avoiding both collisions and barren, coral-deprived regions. To carry out path selection while avoiding collisions, we use monocular image data processed in real time. The proposed system uses a convolutional neural network that takes an image from a forward-facing camera as input and predicts unscaled and relative path changes. The network is trained to encode our desired obstacle-avoidance and reef-exploration objectives via supervised learning from human-labeled data. The predictions from the network are transformed into absolute path changes via a combination of a temporally-smoothed proportional controller for heading targets and a low-level motor controller. This system enables safe and autonomous coral reef navigation in underwater environments. We validate our approach using an untethered and fully autonomous robot swimming through coral reef in the open ocean. Our robot successfully traverses 1000 m of the ocean floor collision-free while collecting close-up footage of coral reefs.

We present an approach to learning control policies for physical robots that achieves high efficiency by adjusting existing policies that have been learned on similar source systems, such as a similar robot with different physical parameters, or an approximate dynamics model simulator. This can be viewed as calibrating a policy learned on a source system, to match a desired behaviour in similar target systems. Our approach assumes that the trajectories described by the source robot are feasible on the target robot. By making this assumption, we only need to learn a mapping from the source robot state and action spaces to the target robot action space, which we call a policy adjustment model. We demonstrate our approach in simulation in the cart-pole balancing task and a two link double pendulum. We also validate our approach with a physical cart-pole system, where we adjust a learned policy under changes to the weight of the pole.

We present a robust multi-robot convoying approach that relies on visual detection of the leading agent, thus enabling target following in unstructured 3-D environments. Our method is based on the idea of tracking-by-detection, which interleaves efficient model-based object detection with temporal filtering of image-based bounding box estimation. This approach has the important advantage of mitigating tracking drift (i.e. drifting away from the target object), which is a common symptom of model-free trackers and is detrimental to sustained convoying in practice. To illustrate our solution, we collected extensive footage of an underwater robot in ocean settings, and hand-annotated its location in each frame. Based on this dataset, we present an empirical comparison of multiple tracker variants, including the use of several convolutional neural networks, both with and without recurrent connections, as well as frequency-based model-free trackers. We also demonstrate the practicality of this tracking-by-detection strategy in real-world scenarios by successfully controlling a legged underwater robot in five degrees of freedom to follow another robot's independent motion.

We present an end-to-end framework for realizing fully automated gait learning for a complex underwater legged robot. Using this framework, we demonstrate that a hexapod flipper-propelled robot can learn task-specific control policies purely from experience data. Our method couples a state-of-the-art policy search technique with a family of periodic low-level controls that are well suited for underwater propulsion. We demonstrate the practical efficacy of tabula rasa learning, that is, learning without the use of any prior knowledge, of policies for a six-legged swimmer to carry out a variety of acrobatic maneuvers in three dimensional space. We also demonstrate informed learning that relies on simulated experience from a realistic simulator. In numerous cases, novel emergent gait behaviors have arisen from learning, such as the use of one stationary flipper to create drag while another oscillates to create thrust. Similar effective results have been demonstrated in under-actuated configurations, where as few as two flippers are used to maneuver the robot to a desired pose, or through an acrobatic motion such as a corkscrew. The success of our learning framework is assessed both in simulation and in the field using an underwater swimming robot.

We study the problem of distributing a single global task between a group of heterogeneous robots. We view this problem as a fair division game. In this setting, every robot defines a preference function over parts of the task according to its sensing and motion capabilities. These preferences are described by density functions over the task. With such interpretation, we want to find an allocation of the global task that maximizes the probability of task completion. We first formulate the task distribution problem as a fair subdivision problem and provide a centralized algorithm to compute the allocations for each robot. We provide a complexity analysis and computational results of the algorithm.

In this paper we describe a heterogeneous multi-robot system for assisting scientists in environmental monitoring tasks, such as the inspection of marine ecosystems. This team of robots is comprised of a fixed-wing aerial vehicle, an autonomous airboat, and an agile legged underwater robot. These robots interact with off-site scientists and operate in a hierarchical structure to autonomously collect visual footage of interesting underwater regions, from multiple scales and mediums. We discuss organizational and scheduling complexities associated with multi-robot experiments in a field robotics setting. We also present results from our field trials, where we demonstrated the use of this heterogeneous robot team to achieve multi-domain monitoring of coral reefs, based on real-time interaction with a remotely-located marine biologist.