Safe exploration is critical for using reinforcement learning (RL) in risk-sensitive environments. Recent work learns risk measures which measure the probability of violating constraints, which can then be used to enable safety. However, learning such risk measures requires significant interaction with the environment, resulting in excessive constraint violations during learning. Furthermore, these measures are not easily transferable to new environments. We cast safe exploration as an offline meta-RL problem, where the objective is to leverage examples of safe and unsafe behavior across a range of environments to quickly adapt learned risk measures to a new environment with previously unseen dynamics. We then propose MEta-learning for Safe Adaptation (MESA), an approach for meta-learning a risk measure for safe RL. Simulation experiments across 5 continuous control domains suggest that MESA can leverage offline data from a range of different environments to reduce constraint violations in unseen environments by up to a factor of 2 while maintaining task performance.

In industrial part kitting, 3D objects are inserted into cavities for transportation or subsequent assembly. Kitting is a critical step as it can decrease downstream processing and handling times and enable lower storage and shipping costs. We present Kit-Net, a framework for kitting previously unseen 3D objects into cavities given depth images of both the target cavity and an object held by a gripper in an unknown initial orientation. Kit-Net uses self-supervised deep learning and data augmentation to train a convolutional neural network (CNN) to robustly estimate 3D rotations between objects and matching concave or convex cavities using a large training dataset of simulated depth images pairs. Kit-Net then uses the trained CNN to implement a controller to orient and position novel objects for insertion into novel prismatic and conformal 3D cavities. Experiments in simulation suggest that Kit-Net can orient objects to have a 98.9% average intersection volume between the object mesh and that of the target cavity. Physical experiments with industrial objects succeed in 18% of trials using a baseline method and in 63% of trials with Kit-Net. Video, code, and data are available at https://github.com/BerkeleyAutomation/Kit-Net.

Reinforcement learning (RL) algorithms have shown impressive success in exploring high-dimensional environments to learn complex, long-horizon tasks, but can often exhibit unsafe behaviors and require extensive environment interaction when exploration is unconstrained. A promising strategy for safe learning in dynamically uncertain environments is requiring that the agent can robustly return to states where task success (and therefore safety) can be guaranteed. While this approach has been successful in low-dimensions, enforcing this constraint in environments with high-dimensional state spaces, such as images, is challenging. We present Latent Space Safe Sets (LS3), which extends this strategy to iterative, long-horizon tasks with image observations by using suboptimal demonstrations and a learned dynamics model to restrict exploration to the neighborhood of a learned Safe Set where task completion is likely. We evaluate LS3 on 4 domains, including a challenging sequential pushing task in simulation and a physical cable routing task. We find that LS3 can use prior task successes to restrict exploration and learn more efficiently than prior algorithms while satisfying constraints. See https://tinyurl.com/latent-ss for code and supplementary material.

Robot manipulation for untangling 1D deformable structures such as ropes, cables, and wires is challenging due to their infinite dimensional configuration space, complex dynamics, and tendency to self-occlude. Analytical controllers often fail in the presence of dense configurations, due to the difficulty of grasping between adjacent cable segments. We present two algorithms that enhance robust cable untangling, LOKI and SPiDERMan, which operate alongside HULK, a high-level planner from prior work. LOKI uses a learned model of manipulation features to refine a coarse grasp keypoint prediction to a precise, optimized location and orientation, while SPiDERMan uses a learned model to sense task progress and apply recovery actions. We evaluate these algorithms in physical cable untangling experiments with 336 knots and over 1500 actions on real cables using the da Vinci surgical robot. We find that the combination of HULK, LOKI, and SPiDERMan is able to untangle dense overhand, figure-eight, double-overhand, square, bowline, granny, stevedore, and triple-overhand knots. The composition of these methods successfully untangles a cable from a dense initial configuration in 68.3% of 60 physical experiments and achieves 50% higher success rates than baselines from prior work. Supplementary material, code, and videos can be found at https://tinyurl.com/rssuntangling.

The difficulty in specifying rewards for many real-world problems has led to an increased focus on learning rewards from human feedback, such as demonstrations. However, there are often many different reward functions that explain the human feedback, leaving agents with uncertainty over what the true reward function is. While most policy optimization approaches handle this uncertainty by optimizing for expected performance, many applications demand risk-averse behavior. We derive a novel policy gradient-style robust optimization approach, PG-BROIL, that optimizes a soft-robust objective that balances expected performance and risk. To the best of our knowledge, PG-BROIL is the first policy optimization algorithm robust to a distribution of reward hypotheses which can scale to continuous MDPs. Results suggest that PG-BROIL can produce a family of behaviors ranging from risk-neutral to risk-averse and outperforms state-of-the-art imitation learning algorithms when learning from ambiguous demonstrations by hedging against uncertainty, rather than seeking to uniquely identify the demonstrator's reward function.

Disentangling two or more cables requires many steps to remove crossings between and within cables. We formalize the problem of disentangling multiple cables and present an algorithm, Iterative Reduction Of Non-planar Multiple cAble kNots (IRON-MAN), that outputs robot actions to remove crossings from multi-cable knotted structures. We instantiate this algorithm with a learned perception system, inspired by prior work in single-cable untying that given an image input, can disentangle two-cable twists, three-cable braids, and knots of two or three cables, such as overhand, square, carrick bend, sheet bend, crown, and fisherman's knots. IRON-MAN keeps track of task-relevant keypoints corresponding to target cable endpoints and crossings and iteratively disentangles the cables by identifying and undoing crossings that are critical to knot structure. Using a da Vinci surgical robot, we experimentally evaluate the effectiveness of IRON-MAN on untangling multi-cable knots of types that appear in the training data, as well as generalizing to novel classes of multi-cable knots. Results suggest that IRON-MAN is effective in disentangling knots involving up to three cables with 80.5% success and generalizing to knot types that are not present during training, with cables of both distinct or identical colors.

Corrective interventions while a robot is learning to automate a task provide an intuitive method for a human supervisor to assist the robot and convey information about desired behavior. However, these interventions can impose significant burden on a human supervisor, as each intervention interrupts other work the human is doing, incurs latency with each context switch between supervisor and autonomous control, and requires time to perform. We present LazyDAgger, which extends the interactive imitation learning (IL) algorithm SafeDAgger to reduce context switches between supervisor and autonomous control. We find that LazyDAgger improves the performance and robustness of the learned policy during both learning and execution while limiting burden on the supervisor. Simulation experiments suggest that LazyDAgger can reduce context switches by an average of 60% over SafeDAgger on 3 continuous control tasks while maintaining state-of-the-art policy performance. In physical fabric manipulation experiments with an ABB YuMi robot, LazyDAgger reduces context switches by 60% while achieving a 60% higher success rate than SafeDAgger at execution time.

Robotic fabric manipulation has applications in home robotics, textiles, senior care and surgery. Existing fabric manipulation techniques, however, are designed for specific tasks, making it difficult to generalize across different but related tasks. We build upon the Visual Foresight framework to learn fabric dynamics that can be efficiently reused to accomplish different sequential fabric manipulation tasks with a single goal-conditioned policy. We extend our earlier work on VisuoSpatial Foresight (VSF), which learns visual dynamics on domain randomized RGB images and depth maps simultaneously and completely in simulation. In this earlier work, we evaluated VSF on multi-step fabric smoothing and folding tasks against 5 baseline methods in simulation and on the da Vinci Research Kit (dVRK) surgical robot without any demonstrations at train or test time. A key finding was that depth sensing significantly improves performance: RGBD data yields an 80% improvement in fabric folding success rate in simulation over pure RGB data. In this work, we vary 4 components of VSF, including data generation, the choice of visual dynamics model, cost function, and optimization procedure. Results suggest that training visual dynamics models using longer, corner-based actions can improve the efficiency of fabric folding by 76% and enable a physical sequential fabric folding task that VSF could not previously perform with 90% reliability. Code, data, videos, and supplementary material are available at https://sites.google.com/view/fabric-vsf/.

There has been significant recent work on data-driven algorithms for learning general-purpose grasping policies. However, these policies can consistently fail to grasp challenging objects which are significantly out of the distribution of objects in the training data or which have very few high quality grasps. Motivated by such objects, we propose a novel problem setting, Exploratory Grasping, for efficiently discovering reliable grasps on an unknown polyhedral object via sequential grasping, releasing, and toppling. We formalize Exploratory Grasping as a Markov Decision Process, study the theoretical complexity of Exploratory Grasping in the context of reinforcement learning and present an efficient bandit-style algorithm, Bandits for Online Rapid Grasp Exploration Strategy (BORGES), which leverages the structure of the problem to efficiently discover high performing grasps for each object stable pose. BORGES can be used to complement any general-purpose grasping algorithm with any grasp modality (parallel-jaw, suction, multi-fingered, etc) to learn policies for objects in which they exhibit persistent failures. Simulation experiments suggest that BORGES can significantly outperform both general-purpose grasping pipelines and two other online learning algorithms and achieves performance within 5% of the optimal policy within 1000 and 8000 timesteps on average across 46 challenging objects from the Dex-Net adversarial and EGAD! object datasets, respectively. Initial physical experiments suggest that BORGES can improve grasp success rate by 45% over a Dex-Net baseline with just 200 grasp attempts in the real world. See https://tinyurl.com/exp-grasping for supplementary material and videos.

The ability of robots to grasp novel objects has industry applications in e-commerce order fulfillment and home service. Data-driven grasping policies have achieved success in learning general strategies for grasping arbitrary objects. However, these approaches can fail to grasp objects which have complex geometry or are significantly outside of the training distribution. We present a Thompson sampling algorithm that learns to grasp a given object with unknown geometry using online experience. The algorithm leverages learned priors from the Dexterity Network robot grasp planner to guide grasp exploration and provide probabilistic estimates of grasp success for each stable pose of the novel object. We find that seeding the policy with the Dex-Net prior allows it to more efficiently find robust grasps on these objects. Experiments suggest that the best learned policy attains an average total reward 64.5% higher than a greedy baseline and achieves within 5.7% of an oracle baseline when evaluated over 300,000 training runs across a set of 3000 object poses.