Learning meaningful visual representations in an embedding space can facilitate generalization in downstream tasks such as action segmentation and imitation. In this paper, we learn a motion-centric representation of surgical video demonstrations by grouping them into action segments/subgoals/options in a semi-supervised manner. We present Motion2Vec, an algorithm that learns a deep embedding feature space from video observations by minimizing a metric learning loss in a Siamese network: images from the same action segment are pulled together while pushed away from randomly sampled images of other segments, while respecting the temporal ordering of the images. The embeddings are iteratively segmented with a recurrent neural network for a given parametrization of the embedding space after pre-training the Siamese network. We only use a small set of labeled video segments to semantically align the embedding space and assign pseudo-labels to the remaining unlabeled data by inference on the learned model parameters. We demonstrate the use of this representation to imitate surgical suturing kinematic motions from publicly available videos of the JIGSAWS dataset. Results give 85.5% segmentation accuracy on average suggesting performance improvement over several state-of-the-art baselines, while kinematic pose imitation gives 0.94 centimeter error in position per observation on the test set. Videos, code and data are available at: https://sites.google.com/view/motion2vec.

Sequential pulling policies to flatten and smooth fabrics have applications from surgery to manufacturing to home tasks such as bed making and folding clothes. Due to the complexity of fabric states and dynamics, we apply deep imitation learning to learn policies that, given color (RGB), depth (D), or combined color-depth (RGBD) images of a rectangular fabric sample, estimate pick points and pull vectors to spread the fabric to maximize coverage. To generate data, we develop a fabric simulator and an algorithmic supervisor that has access to complete state information. We train policies in simulation using domain randomization and dataset aggregation (DAgger) on three tiers of difficulty in the initial randomized configuration. We present results comparing five baseline policies to learned policies and report systematic comparisons of RGB vs D vs RGBD images as inputs. In simulation, learned policies achieve comparable or superior performance to analytic baselines. In 180 physical experiments with the da Vinci Research Kit (dVRK) surgical robot, RGBD policies trained in simulation attain coverage of 83% to 95% depending on difficulty tier, suggesting that effective fabric smoothing policies can be learned from an algorithmic supervisor and that depth sensing is a valuable addition to color alone. Supplementary material is available at https://sites.google.com/view/fabric-smoothing.

The growing demand of industrial, automotive and service robots presents a challenge to the centralized Cloud Robotics model in terms of privacy, security, latency, bandwidth, and reliability. In this paper, we present a `Fog Robotics' approach to deep robot learning that distributes compute, storage and networking resources between the Cloud and the Edge in a federated manner. Deep models are trained on non-private (public) synthetic images in the Cloud; the models are adapted to the private real images of the environment at the Edge within a trusted network and subsequently, deployed as a service for low-latency and secure inference/prediction for other robots in the network. We apply this approach to surface decluttering, where a mobile robot picks and sorts objects from a cluttered floor by learning a deep object recognition and a grasp planning model. Experiments suggest that Fog Robotics can improve performance by sim-to-real domain adaptation in comparison to exclusively using Cloud or Edge resources, while reducing the inference cycle time by 4× to successfully declutter 86% of objects over 213 attempts.

Cloud Robotics is a paradigm where multiple robots are connected to cloud services via Internet to access “unlimited” computation power, at the cost of network communication. However, due to limitations such as network latency and variability, it is difficult to control dynamic, human compliant service robots directly from the cloud. In this work, we combine cloud robotics with an agile edge device to build a Fog Robotic system by leveraging an asynchronous protocol with a “heartbeat” signal. We use the system to enable robust teleoperation of a dynamic self-balancing robot from the cloud. We use the system to pick up boxes from static locations, a task commonly performed in warehouse logistics. To make cloud teleoperation more intuitive and efficient, we program a cloud-based image based visual servoing (IBVS) module to automatically assist the cloud teleoperator during the object pickups. Visual feedbacks, including apriltag recognition and tracking, are performed in the cloud to emulate a Fog Robotic object recognition system for IBVS. We demonstrate the feasibility of a dynamic real-time automation system using this cloud-edge hybrid design, which opens up possibilities of deploying dynamic robotic control with deep-learning recognition systems in Fog Robotics. Finally, we show that Fog Robotics enables the self-balancing service robot to pick up a box automatically from a person under unstructured environments.

On-policy imitation learning algorithms such as Dagger evolve a robot control policy by executing it, measuring performance (loss), obtaining corrective feedback from a supervisor, and generating the next policy. As the loss between iterations can vary unpredictably, a fundamental question is under what conditions this process will eventually achieve a converged policy. If one assumes the underlying trajectory distribution is static (stationary), it is possible to prove convergence for Dagger. Cheng and Boots (2018) consider the more realistic model for robotics where the underlying trajectory distribution, which is a function of the policy, is dynamic and show that it is possible to prove convergence when a condition on the rate of change of the trajectory distributions is satisfied. In this paper, we reframe that result using dynamic regret theory from the field of Online Optimization to prove convergence to locally optimal policies for Dagger, Imitation Gradient, and Multiple Imitation Gradient. These results inspire a new algorithm, Adaptive On-Policy Regularization (AOR), that ensures the conditions for convergence. We present simulation results with cart-pole balancing and walker locomotion benchmarks that suggest AOR can significantly decrease dynamic regret and chattering. To our knowledge, this the first application of dynamic regret theory to imitation learning.

Generalizing manipulation skills to new situations requires extracting invariant patterns from demonstrations. For example, the robot needs to understand the demonstrations at a higher level while being invariant to the appearance of the objects, geometric aspects of objects such as its position, size, orientation and viewpoint of the observer in the demonstrations. In this paper, we propose an algorithm that learns a joint probability density function of the demonstrations with invariant formulations of hidden semi-Markov models to extract invariant segments (also called sub-goals or options), and smoothly follow the generated sequence of states with a linear quadratic tracking controller. The algorithm takes as input the demonstrations observed with respect to different coordinate systems describing virtual landmarks or objects of interest, and adapts the segments according to the environmental changes in a systematic manner. We present variants of this algorithm in latent space with low-rank covariance decompositions, semi-tied covariances, and non-parametric online estimation of model parameters under small variance asymptotics; yielding considerably low sample and model complexity for acquiring new manipulation skills. The algorithm allows a Baxter robot to learn a pick-and-place task while avoiding a movable obstacle based on only 4 kinesthetic demonstrations.

Performing remote manipulation tasks by tele-operation with limited bandwidth, communication delays and environmental differences is a challenging problem. In this paper, we learn a task-parameterized generative model from the teleoperator demonstrations using a hidden semi-Markov model that provides assistance in performing remote manipulation tasks. We present a probabilistic formulation to capture the intention of the teleoperator, and subsequently assist the teleoperator by time-independent shared control and/or time-dependent autonomous control formulations of the model. In the shared control mode, the model corrects the remote arm movement based on the current state of the teleoperator; whereas in the autonomous control mode, the model generates the movement of the remote arm for autonomous task execution. We show the formulation of the model with virtual fixtures and provide comparisons to benchmark our approach. Teleoperation experiments with the Baxter robot for reaching a movable target and opening a valve reveal that the proposed methodology improves the performance of the teleoperator and caters for environmental differences in performing remote manipulation tasks.

We consider the problem of incrementally learning different strategies of performing a complex sequential task from multiple demonstrations of an expert or a set of experts. While the task is the same, each expert differs in his/her way of performing it. We assume that this variety across experts' demonstration is due to the fact that each expert/strategy is driven by a different reward function, where reward function is expressed as a linear combination of a set of known features. Consequently, we can learn all the expert strategies by forming a convex set of optimal deterministic policies, from which one can match any unseen expert strategy drawn from this set. Instead of learning from scratch every optimal policy in this set, the learner transfers knowledge from the set of learned policies to bootstrap its search for new optimal policy. We demonstrate our approach on a simulated mini-golf task where the 7 degrees of freedom Barrett WAM robot arm learns to sequentially putt on different holes in accordance with the playing strategies of the expert.

The sophistication of computer malware is becoming a serious threat to the information technology infrastructure, which is the backbone of modern e-commerce systems. We, therefore, advocate the need for developing sophisticated, efficient, and accurate malware classification techniques that can detect a malware on the first day of its launch -- commonly known as "zero-day malware detection". To this end, we present a new technique, IMAD, that can not only identify zero-day malware without any apriori knowledge but can also detect a malicious process while it is executing (in-execution detection). The capability of in-execution malware detection empowers an operating system to immediately kill it before it can cause any significant damage. IMAD is a realtime, dynamic, efficient, in-execution zero-day malware detection scheme, which analyzes the system call sequence of a process to classify it as malicious or benign. We use Genetic Algorithm to optimize system parameters of our scheme. The evolutionary algorithm is evaluated on real world synthetic data extracted from a Linux system. The results of our experiments show that IMAD achieves more than 90% accuracy in classifying in-execution processes as benign or malicious. Moreover, our scheme can classify approximately 50% of malicious processes within first 20% of their system calls.