arXiv subjects

Ken Goldberg

Publications and source records attributed to Ken Goldberg.

191 records · Page 11Linked to original sources

arXiv2017

BoostClean: Automated Error Detection and Repair for Machine Learning

Predictive models based on machine learning can be highly sensitive to data error. Training data are often combined with a variety of different sources, each susceptible to different types of inconsistencies, and new data streams during prediction time, the model may encounter previously unseen inconsistencies. An important class of such inconsistencies is domain value violations that occur when an attribute value is outside of an allowed domain. We explore automatically detecting and repairing such violations by leveraging the often available clean test labels to determine whether a given detection and repair combination will improve model accuracy. We present BoostClean which automatically selects an ensemble of error detection and repair combinations using statistical boosting. BoostClean selects this ensemble from an extensible library that is pre-populated general detection functions, including a novel detector based on the Word2Vec deep learning model, which detects errors across a diverse set of domains. Our evaluation on a collection of 12 datasets from Kaggle, the UCI repository, real-world data analyses, and production datasets that show that Boost- Clean can increase absolute prediction accuracy by up to 9% over the best non-ensembled alternatives. Our optimizations including parallelism, materialization, and indexing techniques show a 22.2x end-to-end speedup on a 16-core machine.

cs.DB↗

arXiv2017

DDCO: Discovery of Deep Continuous Options for Robot Learning from Demonstrations

An option is a short-term skill consisting of a control policy for a specified region of the state space, and a termination condition recognizing leaving that region. In prior work, we proposed an algorithm called Deep Discovery of Options (DDO) to discover options to accelerate reinforcement learning in Atari games. This paper studies an extension to robot imitation learning, called Discovery of Deep Continuous Options (DDCO), where low-level continuous control skills parametrized by deep neural networks are learned from demonstrations. We extend DDO with: (1) a hybrid categorical-continuous distribution model to parametrize high-level policies that can invoke discrete options as well continuous control actions, and (2) a cross-validation method that relaxes DDO's requirement that users specify the number of options to be discovered. We evaluate DDCO in simulation of a 3-link robot in the vertical plane pushing a block with friction and gravity, and in two physical experiments on the da Vinci surgical robot, needle insertion where a needle is grasped and inserted into a silicone tissue phantom, and needle bin picking where needles and pins are grasped from a pile and categorized into bins. In the 3-link arm simulation, results suggest that DDCO can take 3x fewer demonstrations to achieve the same reward compared to a baseline imitation learning approach. In the needle insertion task, DDCO was successful 8/10 times compared to the next most accurate imitation learning baseline 6/10. In the surgical bin picking task, the learned policy successfully grasps a single object in 66 out of 99 attempted grasps, and in all but one case successfully recovered from failed grasps by retrying a second time.

cs.RO↗

arXiv2017

DART: Noise Injection for Robust Imitation Learning

One approach to Imitation Learning is Behavior Cloning, in which a robot observes a supervisor and infers a control policy. A known problem with this "off-policy" approach is that the robot's errors compound when drifting away from the supervisor's demonstrations. On-policy, techniques alleviate this by iteratively collecting corrective actions for the current robot policy. However, these techniques can be tedious for human supervisors, add significant computation burden, and may visit dangerous states during training. We propose an off-policy approach that injects noise into the supervisor's policy while demonstrating. This forces the supervisor to demonstrate how to recover from errors. We propose a new algorithm, DART (Disturbances for Augmenting Robot Trajectories), that collects demonstrations with injected noise, and optimizes the noise level to approximate the error of the robot's trained policy during data collection. We compare DART with DAgger and Behavior Cloning in two domains: in simulation with an algorithmic supervisor on the MuJoCo tasks (Walker, Humanoid, Hopper, Half-Cheetah) and in physical experiments with human supervisors training a Toyota HSR robot to perform grasping in clutter. For high dimensional tasks like Humanoid, DART can be up to $3x$ faster in computation time and only decreases the supervisor's cumulative reward by $5\%$ during training, whereas DAgger executes policies that have $80\%$ less cumulative reward than the supervisor. On the grasping in clutter task, DART obtains on average a $62\%$ performance increase over Behavior Cloning.

cs.LG↗

arXiv2017

Multi-Level Discovery of Deep Options

Augmenting an agent's control with useful higher-level behaviors called options can greatly reduce the sample complexity of reinforcement learning, but manually designing options is infeasible in high-dimensional and abstract state spaces. While recent work has proposed several techniques for automated option discovery, they do not scale to multi-level hierarchies and to expressive representations such as deep networks. We present Discovery of Deep Options (DDO), a policy-gradient algorithm that discovers parametrized options from a set of demonstration trajectories, and can be used recursively to discover additional levels of the hierarchy. The scalability of our approach to multi-level hierarchies stems from the decoupling of low-level option discovery from high-level meta-control policy learning, facilitated by under-parametrization of the high level. We demonstrate that using the discovered options to augment the action space of Deep Q-Network agents can accelerate learning by guiding exploration in tasks where random actions are unlikely to reach valuable states. We show that DDO is effective in adding options that accelerate learning in 4 out of 5 Atari RAM environments chosen in our experiments. We also show that DDO can discover structure in robot-assisted surgical videos and kinematics that match expert annotation with 72% accuracy.

cs.LG↗

arXiv2017

Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics

To reduce data collection time for deep learning of robust robotic grasp plans, we explore training from a synthetic dataset of 6.7 million point clouds, grasps, and analytic grasp metrics generated from thousands of 3D models from Dex-Net 1.0 in randomized poses on a table. We use the resulting dataset, Dex-Net 2.0, to train a Grasp Quality Convolutional Neural Network (GQ-CNN) model that rapidly predicts the probability of success of grasps from depth images, where grasps are specified as the planar position, angle, and depth of a gripper relative to an RGB-D sensor. Experiments with over 1,000 trials on an ABB YuMi comparing grasp planning methods on singulated objects suggest that a GQ-CNN trained with only synthetic data from Dex-Net 2.0 can be used to plan grasps in 0.8sec with a success rate of 93% on eight known objects with adversarial geometry and is 3x faster than registering point clouds to a precomputed dataset of objects and indexing grasps. The Dex-Net 2.0 grasp planner also has the highest success rate on a dataset of 10 novel rigid objects and achieves 99% precision (one false positive out of 69 grasps classified as robust) on a dataset of 40 novel household objects, some of which are articulated or deformable. Code, datasets, videos, and supplementary material are available at http://berkeleyautomation.github.io/dex-net .

cs.RO↗

arXiv2017

Comparing Human-Centric and Robot-Centric Sampling for Robot Deep Learning from Demonstrations

Motivated by recent advances in Deep Learning for robot control, this paper considers two learning algorithms in terms of how they acquire demonstrations. "Human-Centric" (HC) sampling is the standard supervised learning algorithm, where a human supervisor demonstrates the task by teleoperating the robot to provide trajectories consisting of state-control pairs. "Robot-Centric" (RC) sampling is an increasingly popular alternative used in algorithms such as DAgger, where a human supervisor observes the robot executing a learned policy and provides corrective control labels for each state visited. RC sampling can be challenging for human supervisors and prone to mislabeling. RC sampling can also induce error in policy performance because it repeatedly visits areas of the state space that are harder to learn. Although policies learned with RC sampling can be superior to HC sampling for standard learning models such as linear SVMs, policies learned with HC sampling may be comparable with highly-expressive learning models such as deep learning and hyper-parametric decision trees, which have little model error. We compare HC and RC using a grid world and a physical robot singulation task, where in the latter the input is a binary image of a connected set of objects on a planar worksurface and the policy generates a motion of the gripper to separate one object from the rest. We observe in simulation that for linear SVMs, policies learned with RC outperformed those learned with HC but that with deep models this advantage disappears. We also find that with RC, the corrective control labels provided by humans can be highly inconsistent. We prove there exists a class of examples where in the limit, HC is guaranteed to converge to an optimal policy while RC may fail to converge.

cs.RO↗

arXiv2016

Toward a Science of Autonomy for Physical Systems: Paths

An Autonomous Physical System (APS) will be expected to reliably and independently evaluate, execute, and achieve goals while respecting surrounding rules, laws, or conventions. In doing so, an APS must rely on a broad spectrum of dynamic, complex, and often imprecise information about its surroundings, the task it is to perform, and its own sensors and actuators. For example, cleaning in a home or commercial setting requires the ability to perceive, grasp, and manipulate many physical objects, the ability to reliably perform a variety of subtasks such as washing, folding, and stacking, and knowledge about local conventions such as how objects are classified and where they should be stored. The information required for reliable autonomous operation may come from external sources and from the robot's own sensor observations or in the form of direct instruction by a trainer. Similar considerations apply across many domains - construction, manufacturing, in-home assistance, and healthcare. For example, surgeons spend many years learning about physiology and anatomy before they touch a patient. They then perform roughly 1000 surgeries under the tutelage of an expert surgeon, and they practice basic maneuvers such as suture tying thousands of times outside the operating room. All of these elements come together to achieve expertise at this task. Endowing a system with robust autonomy by traditional programming methods has thus far had limited success. Several promising new paths to acquiring and processing such data are emerging. This white paper outlines three promising research directions for enabling an APS to learn the physical and information skills necessary to perform tasks with independence and flexibility: Deep Reinforcement Learning, Human-Robot Interaction, and Cloud Robotics.

cs.CY↗

arXiv2016

HIRL: Hierarchical Inverse Reinforcement Learning for Long-Horizon Tasks with Delayed Rewards

Reinforcement Learning (RL) struggles in problems with delayed rewards, and one approach is to segment the task into sub-tasks with incremental rewards. We propose a framework called Hierarchical Inverse Reinforcement Learning (HIRL), which is a model for learning sub-task structure from demonstrations. HIRL decomposes the task into sub-tasks based on transitions that are consistent across demonstrations. These transitions are defined as changes in local linearity w.r.t to a kernel function. Then, HIRL uses the inferred structure to learn reward functions local to the sub-tasks but also handle any global dependencies such as sequentiality. We have evaluated HIRL on several standard RL benchmarks: Parallel Parking with noisy dynamics, Two-Link Pendulum, 2D Noisy Motion Planning, and a Pinball environment. In the parallel parking task, we find that rewards constructed with HIRL converge to a policy with an 80% success rate in 32% fewer time-steps than those constructed with Maximum Entropy Inverse RL (MaxEnt IRL), and with partial state observation, the policies learned with IRL fail to achieve this accuracy while HIRL still converges. We further find that that the rewards learned with HIRL are robust to environment noise where they can tolerate 1 stdev. of random perturbation in the poses in the environment obstacles while maintaining roughly the same convergence rate. We find that HIRL rewards can converge up-to 6x faster than rewards constructed with IRL.

cs.RO↗

arXiv2016

ActiveClean: Interactive Data Cleaning While Learning Convex Loss Models

Data cleaning is often an important step to ensure that predictive models, such as regression and classification, are not affected by systematic errors such as inconsistent, out-of-date, or outlier data. Identifying dirty data is often a manual and iterative process, and can be challenging on large datasets. However, many data cleaning workflows can introduce subtle biases into the training processes due to violation of independence assumptions. We propose ActiveClean, a progressive cleaning approach where the model is updated incrementally instead of re-training and can guarantee accuracy on partially cleaned data. ActiveClean supports a popular class of models called convex loss models (e.g., linear regression and SVMs). ActiveClean also leverages the structure of a user's model to prioritize cleaning those records likely to affect the results. We evaluate ActiveClean on five real-world datasets UCI Adult, UCI EEG, MNIST, Dollars For Docs, and WorldBank with both real and synthetic errors. Our results suggest that our proposed optimizations can improve model accuracy by up-to 2.5x for the same amount of data cleaned. Furthermore for a fixed cleaning budget and on all real dirty datasets, ActiveClean returns more accurate models than uniform sampling and Active Learning.

cs.DB↗

arXiv2015

Stale View Cleaning: Getting Fresh Answers from Stale Materialized Views

Materialized views (MVs), stored pre-computed results, are widely used to facilitate fast queries on large datasets. When new records arrive at a high rate, it is infeasible to continuously update (maintain) MVs and a common solution is to defer maintenance by batching updates together. Between batches the MVs become increasingly stale with incorrect, missing, and superfluous rows leading to increasingly inaccurate query results. We propose Stale View Cleaning (SVC) which addresses this problem from a data cleaning perspective. In SVC, we efficiently clean a sample of rows from a stale MV, and use the clean sample to estimate aggregate query results. While approximate, the estimated query results reflect the most recent data. As sampling can be sensitive to long-tailed distributions, we further explore an outlier indexing technique to give increased accuracy when the data distributions are skewed. SVC complements existing deferred maintenance approaches by giving accurate and bounded query answers between maintenance. We evaluate our method on a generated dataset from the TPC-D benchmark and a real video distribution application. Experiments confirm our theoretical results: (1) cleaning an MV sample is more efficient than full view maintenance, (2) the estimated results are more accurate than using the stale MV, and (3) SVC is applicable for a wide variety of MVs.

cs.DB↗

arXiv2010

IPIP: A New Approach to Inverse Planning for HDR Brachytherapy by Directly Optimizing Dosimetric Indices

Purpose: Many planning methods for high dose rate (HDR) brachytherapy treatment planning require an iterative approach. A set of computational parameters are hypothesized that will give a dose plan that meets dosimetric criteria. A dose plan is computed using these parameters, and if any dosimetric criteria are not met, the process is iterated until a suitable dose plan is found. In this way, the dose distribution is controlled by abstract parameters. The purpose of this study is to improve HDR brachytherapy planning by developing a new approach that directly optimizes the dose distribution based on dosimetric criteria. Method: We develop Inverse Planning by Integer Program (IPIP), an optimization model for computing HDR brachytherapy dose plans and a fast heuristic for it. We used our heuristic to compute dose plans for 20 anonymized prostate cancer patient image data sets from our clinic database. Dosimetry was evaluated and compared to dosimetric criteria. Results: Dose plans computed from IPIP satisfied all given dosimetric criteria for the target and healthy tissue after a single iteration. The average target coverage was 95%. The average computation time for IPIP was 30.1 seconds on a Intel(R) CoreTM2 Duo CPU 1.67 GHz processor with 3 Gib RAM. Conclusion: IPIP is an HDR brachytherapy planning system that directly incorporates dosimetric criteria. We have demonstrated that IPIP has clinically acceptable performance for the prostate cases and dosimetric criteria used in this study, both in terms of dosimetry and runtime. Further study is required to determine if IPIP performs well for a more general group of patients and dosimetric criteria, including other cancer sites such as GYN.

physics.med-ph↗