IROS 2020 Papers

Utilization of latent space to capture a lower-dimensional representation of a complex dynamics model is explored in this work. The targeted application is of a robotic manipulator executing a complex environment interaction task, in particular, cutting a wooden object. We train two flavours of Variational Autoencoders-standard and Vector-Quantised-to learn the latent space which is then used to infer certain properties of the cutting operation, such as whether the robot is cutting or not, as well as, material and geometry of the object being cut. The two VAE models are evaluated with reconstruction, prediction and a combined reconstruction/prediction decoders. The results demonstrate the expressiveness of the latent space for robotic interaction inference and the competitive prediction performance against recurrent neural networks.

In this paper, we aim to expedite the deployment of challenging manipulation tasks involving both motion and contact wrenches (forces and moments). To this end, we acquire motion and wrench signals from a small set of demonstrations using passive observation. To learn these tasks, we introduce Trajectory parameterized Probabilistic Principal Component Analysis (traPPCA) which compactly re-parameterizes the acquired signals using trajectory information and encodes the signal correlations using Probabilistic Principal Component Analysis (PPCA). Finally, the task is transferred to a robot setup by specifying the robot behavior using a constraint-based task specification and control approach. This framework results in increased robustness of the system against different sources of uncertainty: imprecise sensors, adaptation of the tool, and changes in the execution speed.

The purpose of this study is to make life-sized humanoid robots acquire tool manipulation skills that require complicated force adjustment. The difficulty in acquisition of tool manipulation skills comes from the hardship in physical modeling. Recent research have revealed that deep reinforcement learning (DRL), a model-free approach, performs superior in such tasks. However, DRL in general has a drawback in sample efficiency, and this becomes critical in robot learning especially in life-sized humanoid robots. In this study, we propose an integrated system incorporating DRL method and active learning. Our method also leverages a variety of previous studies on life-sized humanoid robots to overcome the sample efficiency issue. We demonstrated the effectiveness of our proposed system through a hacksaw skill acquisition and a Japanese planer (Kanna) skill acquisition by a life-sized humanoid robot.

Learning-based driving solution, a new branch for autonomous driving, is expected to simplify the modeling of driving by learning the underlying mechanisms from data. To improve the tactical decision-making for learning-based driving solution, we introduce hierarchical behavior and motion planning (HBMP) to explicitly model the behavior in learning-based solution. Due to the coupled action space of behavior and motion, it is challenging to solve HBMP problem using reinforcement learning (RL) for long-horizon driving tasks. We transform HBMP problem by integrating a classical sampling-based motion planner, of which the optimal cost is regarded as the rewards for high-level behavior learning. As a result, this formulation reduces action space and diversifies the rewards without losing the optimality of HBMP. In addition, we propose a sharable representation for input sensory data across simulation platforms and real-world environment, so that models trained in a fast event-based simulator, SUMO, can be used to initialize and accelerate the RL training in a dynamics based simulator, CARLA. Experimental results demonstrate the effectiveness of the method. Besides, the model is successfully transferred to the real-world, validating the generalization capability.

How can we learn representations for planning that are both efficient and flexible? Task and motion planning models are a good candidate, having been very successful in long-horizon planning tasks-however, they've proved challenging for learning, relying mostly on hand-coded representations. We present a framework for learning constraint-based task and motion planning models using gradient descent. Our model observes expert demonstrations of a task and decomposes them into modes-segments which specify a set of constraints on a trajectory optimization problem. We show that our model learns these modes from few demonstrations, that modes can be used to plan flexibly in different environments and to achieve different types of goals, and that the model can recombine these modes in novel ways.

Enabling robots to quickly learn manipulation skills is an important, yet challenging problem. Such manipulation skills should be flexible, e.g., be able adapt to the current workspace configuration. Furthermore, to accomplish complex manipulation tasks, robots should be able to sequence several skills and adapt them to changing situations. In this work, we propose a rapid robot skill-sequencing algorithm, where the skills are encoded by object-centric hidden semi-Markov models. The learned skill models can encode multimodal (temporal and spatial) trajectory distributions. This approach significantly reduces manual modeling efforts, while ensuring a high degree of flexibility and re-usability of learned skills. Given a task goal and a set of generic skills, our framework computes smooth transitions between skill instances. To compute the corresponding optimal end-effector trajectory in task space we rely on Riemannian optimal controller. We demonstrate this approach on a 7 DoF robot arm for industrial assembly tasks.

The capability to detect objects is a core part of autonomous driving. Due to sensor noise and incomplete data, perfectly detecting and localizing every object is infeasible. Therefore, it is important for a detector to provide the amount of uncertainty in each prediction. Providing the autonomous system with reliable uncertainties enables the vehicle to react differently based on the level of uncertainty. Previous work has estimated the uncertainty in a detection by predicting a probability distribution over object bounding boxes. In this work, we propose a method to improve the ability to learn the probability distribution by considering the potential noise in the ground-truth labeled data. Our proposed approach improves not only the accuracy of the learned distribution but also the object detection performance.

Localization is a crucial capability for mobile robots and autonomous cars. In this paper, we address learning an observation model for Monte-Carlo localization using 3D LiDAR data. We propose a novel, neural network-based observation model that computes the expected overlap of two 3D LiDAR scans. The model predicts the overlap and yaw angle offset between the current sensor reading and virtual frames generated from a pre-built map. We integrate this observation model into a Monte-Carlo localization framework and tested it on urban datasets collected with a car in different seasons. The experiments presented in this paper illustrate that our method can reliably localize a vehicle in typical urban environments. We furthermore provide comparisons to a beam-endpoint and a histogram-based method indicating a superior global localization performance of our method with fewer particles.

Sampling-based motion planners (SBMP) are commonly used to generate motion plans by incrementally constructing a search tree through a robot's configuration space. For high degree-of-freedom systems, sampling is often done in a lower-dimensional space, with a steering function responsible for local planning in the higher-dimensional configuration space. However, for highly-redundant systems with complex kinematics, this approach is problematic due to the high computational cost of evaluating the steering function, especially in cluttered environments. Therefore, having an efficient, informed sampler becomes critical to online robot operation. In this study, we develop a learning-based approach with policy improvement to compute an optimal sampling distribution for use in SBMPs. Motivated by the challenge of whole-body planning for a 31 degree-of-freedom mobile robot built by the Toyota Research Institute, we combine our learning-based approach with classical graph-search to obtain a constrained sampling distribution. Over multiple learning iterations, the algorithm learns a probability distribution weighting areas of low-cost and high probability of success, which a graph search algorithm then uses to obtain an optimal sampling distribution for the robot. On challenging motion planning tasks for the robot, we observe significant computational speed-up, fewer edge evaluations, and more efficient paths with minimal computational overhead. We show the efficacy of our approach with a number of experiments in whole-body motion planning.

We present a robot navigation system that uses an imitation learning framework to successfully navigate in complex environments. Our framework takes a pre-built 3D scan of a real environment and trains an agent from pre-generated expert trajectories to navigate to any position given a panoramic view of the goal and the current visual input without relying on map, compass, odometry, or relative position of the target at runtime. Our end-to-end trained agent uses RGB and depth (RGBD) information and can handle large environments (up to 1031m2) across multiple rooms (up to 40) and generalizes to unseen targets. We show that when compared to several baselines our method (1) requires fewer training examples and less training time, (2) reaches the goal location with higher accuracy, and (3) produces better solutions with shorter paths for long-range navigation tasks.

Machines are a long way from robustly solving open-world perception-control tasks, such as first-person view (FPV) aerial navigation. While recent advances in end-to- end Machine Learning, especially Imitation Learning and Reinforcement appear promising, they are constrained by the need of large amounts of difficult-to-collect labeled real- world data. Simulated data, on the other hand, is easy to generate, but generally does not render safe behaviors in diverse real-life scenarios. In this work we propose a novel method for learning robust visuomotor policies for real-world deployment which can be trained purely with simulated data. We develop rich state representations that combine supervised and unsupervised environment data. Our approach takes a cross-modal perspective, where separate modalities correspond to the raw camera data and the system states relevant to the task, such as the relative pose of gates to the drone in the case of drone racing. We feed both data modalities into a novel factored architecture, which learns a joint lowdimensional embedding via Variational Auto Encoders. This compact representation is then fed into a control policy, which we trained using imitation learning with expert trajectories in a simulator. We analyze the rich latent spaces learned with our proposed representations, and show that the use of our cross-modal architecture significantly improves control policy performance as compared to end-to-end learning or purely unsupervised feature extractors. We also present real-world results for drone navigation through gates in different track configurations and environmental conditions. Our proposed method, which runs fully onboard, can successfully generalize the learned representations and policies across simulation and reality, significantly outperforming baseline approaches.

Robots operating in human environments must be careful, when executing their manipulation skills, not to disturb nearby objects. This requires robots to reason about the effect of their manipulation choices by accounting for the support relationships among objects in the scene. Humans do this in part by visually assessing their surroundings and using physics intuition for how likely it is that a particular object can be safely manipulated (i.e., cause no disruption in the rest of the scene). Existing work has shown that deep convolutional neural networks can learn intuitive physics over images generated in simulation and determine the stability of a scene in the real world. In this paper, we extend these physics intuition models to the task of assessing safe object extraction by conditioning the visual images on specific objects in the scene. Our results, in both simulation and real-world settings, show that with our proposed method, physics intuition models can be used to inform a robot of which objects can be safely extracted and from which direction to extract them.

When humans control drones, cars, and robots, we often have some preconceived notion of how our inputs should make the system behave. Existing approaches to teleoperation typically assume a one-size-fits-all approach, where the designers pre-define a mapping between human inputs and robot actions, and every user must adapt to this mapping over repeated interactions. Instead, we propose a personalized method for learning the human's preferred or preconceived mapping from a few robot queries. Given a robot controller, we identify an alignment model that transforms the human's inputs so that the controller's output matches their expectations. We make this approach data-efficient by recognizing that human mappings have strong priors: we expect the input space to be proportional, reversable, and consistent. Incorporating these priors ensures that the robot learns an intuitive mapping from few examples. We test our learning approach in robot manipulation tasks inspired by assistive settings, where each user has different personal preferences and physical capabilities for teleoperating the robot arm. Our simulated and experimental results suggest that learning the mapping between inputs and robot actions improves objective and subjective performance when compared to manually defined alignments or learned alignments without intuitive priors. The supplementary video showing these user studies can be found at: https://youtu.be/rKHka0_48-Q.

This paper introduces an algorithm for discovering implicit and delayed causal relations between events observed by a robot at arbitrary times, with the objective of improving data-efficiency and interpretability of model- based reinforcement learning (RL) techniques. The proposed algorithm initially predicts observations with the Markov assumption, and incrementally introduces new hidden variables to explain and reduce the stochasticity of the observations. The hidden variables are memory units that keep track of pertinent past events. Such events are systematically identified by their information gains. The learned transition and reward models are then used for planning. Experiments on simulated and real robotic tasks show that this method significantly improves over current RL techniques.

In this paper, we approach the challenging problem of motion planning for knot tying. We propose a hierarchical approach in which the top layer produces a topological plan and the bottom layer translates this plan into continuous robot motion. The top layer decomposes a knotting task into sequences of abstract topological actions based on knot theory. The bottom layer translates each of these abstract actions into robot motion trajectories through learned topological motion primitives. To adapt each topological action to the specific rope geometry, the motion primitives take the observed rope configuration as input. We train the motion primitives by imitating human demonstrations and reinforcement learning in simulation. To generalize human demonstrations of simple knots into more complex knots, we observe similarities in the motion strategies of different topological actions and design the neural network structure to exploit such similarities. We demonstrate that our learned motion primitives can be used to efficiently generate motion plans for tying the overhand knot. The motion plan can then be executed on a real robot using visual tracking and Model Predictive Control. We also demonstrate that our learned motion primitives can be composed to tie a more complex pentagram-like knot despite being only trained on human demonstrations of simpler knots.

Being able to quickly adapt to changes in dynamics is paramount in model-based control for object manipulation tasks. In order to influence fast adaptation of the inverse dynamics model's parameters, data efficiency is crucial. Given observed data, a key element to how an optimizer updates model parameters is the loss function. In this work, we propose to apply meta-learning to learn structured, state-dependent loss functions during a meta-training phase. We then replace standard losses with our learned losses during online adaptation tasks. We evaluate our proposed approach on inverse dynamics learning tasks, both in simulation and on real hardware data. In both settings, the structured and state-dependent learned losses improve online adaptation speed, when compared to standard, state-independent loss functions.

Physically soft robots are promising for robotic assembly tasks as they allow stable contacts with the environment. In this study, we propose a novel learning system for soft robotic assembly strategies. We formulate this problem as a reinforcement learning task and design the reward function from human demonstrations. Our key insight is that the failed demonstrations can be used as constraints to avoid failed behaviors. To this end, we developed a teaching device with which humans can intuitively provide various demonstrations. Moreover, we leverage Physically-Consistent Gaussian Mixture Models to clearly assign Gaussian components to the successful and failed trials. We then create the reference trajectories via Gaussian Mixture Regressions, which fit the successful demonstrations while considering the failed ones. Finally, we apply a sample- efficient deep model-based reinforcement learning method to obtain robust strategies with a few interactions. To validate our method, we developed a real-robot experimental system composed of a rigid collaborative robot arm with a compliant wrist and the teaching device. Our results demonstrated that our method learned the assembly strategies with a higher success rate than when using only successful demonstrations.

Planners using accurate models can be effective for accomplishing manipulation tasks in the real world, but are typically highly specialized and require significant fine-tuning to be reliable. Meanwhile, learning is useful for adaptation, but can require a substantial amount of data collection. In this paper, we propose a method that improves the efficiency of sub-optimal planners with approximate but simple and fast models by switching to a model-free policy when unexpected transitions are observed. Unlike previous work, our method specifically addresses when the planner fails due to transition model error by patching with a local policy only where needed. First, we use a sub-optimal model-based planner to perform a task until model failure is detected. Next, we learn a local model-free policy from expert demonstrations to complete the task in regions where the model failed. To show the efficacy of our method, we perform experiments with a shape insertion puzzle and compare our results to both pure planning and imitation learning approaches. We then apply our method to a door opening task. Our experiments demonstrate that our patch-enhanced planner performs more reliably than pure planning and with lower overall sample complexity than pure imitation learning.

For robots to operate robustly in the real world, they should be aware of their uncertainty. However, most methods for object pose estimation return a single point estimate of the object's pose. In this work, we propose two learned methods for estimating a distribution over an object's orientation. Our methods take into account both the inaccuracies in the pose estimation as well as the object symmetries. Our first method, which regresses from deep learned features to an isotropic Bingham distribution, gives the best performance for orientation distribution estimation for non-symmetric objects. Our second method learns to compare deep features and generates a non-parameteric histogram distribution. This method gives the best performance on objects with unknown symmetries, accurately modeling both symmetric and non-symmetric objects, without any requirement of symmetry annotation. We show that both of these methods can be used to augment an existing pose estimator. Our evaluation compares our methods to a large number of baseline approaches for uncertainty estimation across a variety of different types of objects. Code available at https://bokorn.github.io/orientation-distributions/.