We develop a novel policy synthesis algorithm, RMPflow, based on geometrically consistent transformations of Riemannian Motion Policies (RMPs). RMPs are a class of reactive motion policies designed to parameterize non-Euclidean behaviors as dynamical systems in intrinsically nonlinear task spaces. Given a set of RMPs designed for individual tasks, RMPflow can consistently combine these local policies to generate an expressive global policy, while simultaneously exploiting sparse structure for computational efficiency. We study the geometric properties of RMPflow and provide sufficient conditions for stability. Finally, we experimentally demonstrate that accounting for the geometry of task policies can simplify classically difficult problems, such as planning through clutter on high-DOF manipulation systems.
Current methods for estimating force from tactile sensor signals are either inaccurate analytic models or task-specific learned models. In this paper, we explore learning a robust model that maps tactile sensor signals to force. We specifically explore learning a mapping for the SynTouch BioTac sensor via neural networks. We propose a voxelized input feature layer for spatial signals and leverage information about the sensor surface to regularize the loss function. To learn a robust tactile force model that transfers across tasks, we generate ground truth data from three different sources: (1) the BioTac rigidly mounted to a force torque~(FT) sensor, (2) a robot interacting with a ball rigidly attached to the same FT sensor, and (3) through force inference on a planar pushing task by formalizing the mechanics as a system of particles and optimizing over the object motion. A total of 140k samples were collected from the three sources. We achieve a median angular accuracy of 3.5 degrees in predicting force direction (66% improvement over the current state of the art) and a median magnitude accuracy of 0.06 N (93% improvement) on a test dataset. Additionally, we evaluate the learned force model in a force feedback grasp controller performing object lifting and gentle placement. Our results can be found on https://sites.google.com/view/tactile-force.
We present structured domain randomization (SDR), a variant of domain randomization (DR) that takes into account the structure and context of the scene. In contrast to DR, which places objects and distractors randomly according to a uniform probability distribution, SDR places objects and distractors randomly according to probability distributions that arise from the specific problem at hand. In this manner, SDR-generated imagery enables the neural network to take the context around an object into consideration during detection. We demonstrate the power of SDR for the problem of 2D bounding box car detection, achieving competitive results on real data after training only on synthetic data. On the KITTI easy, moderate, and hard tasks, we show that SDR outperforms other approaches to generating synthetic data (VKITTI, Sim 200k, or DR), as well as real data collected in a different domain (BDD100K). Moreover, synthetic SDR data combined with real KITTI data outperforms real KITTI data alone.
Using synthetic data for training deep neural networks for robotic manipulation holds the promise of an almost unlimited amount of pre-labeled training data, generated safely out of harm's way. One of the key challenges of synthetic data, to date, has been to bridge the so-called reality gap, so that networks trained on synthetic data operate correctly when exposed to real-world data. We explore the reality gap in the context of 6-DoF pose estimation of known objects from a single RGB image. We show that for this problem the reality gap can be successfully spanned by a simple combination of domain randomized and photorealistic data. Using synthetic data generated in this manner, we introduce a one-shot deep neural network that is able to perform competitively against a state-of-the-art network trained on a combination of real and synthetic data. To our knowledge, this is the first deep network trained only on synthetic data that is able to achieve state-of-the-art performance on 6-DoF object pose estimation. Our network also generalizes better to novel environments including extreme lighting conditions, for which we show qualitative results. Using this network we demonstrate a real-time system estimating object poses with sufficient accuracy for real-world semantic grasping of known household objects in clutter by a real robot.
We introduce the Riemannian Motion Policy (RMP), a new mathematical object for modular motion generation. An RMP is a second-order dynamical system (acceleration field or motion policy) coupled with a corresponding Riemannian metric. The motion policy maps positions and velocities to accelerations, while the metric captures the directions in the space important to the policy. We show that RMPs provide a straightforward and convenient method for combining multiple motion policies and transforming such policies from one space (such as the task space) to another (such as the configuration space) in geometrically consistent ways. The operators we derive for these combinations and transformations are provably optimal, have linearity properties making them agnostic to the order of application, and are strongly analogous to the covariant transformations of natural gradients popular in the machine learning literature. The RMP framework enables the fusion of motion policies from different motion generation paradigms, such as dynamical systems, dynamic movement primitives (DMPs), optimal control, operational space control, nonlinear reactive controllers, motion optimization, and model predictive control (MPC), thus unifying these disparate techniques from the literature. RMPs are easy to implement and manipulate, facilitate controller design, simplify handling of joint limits, and clarify a number of open questions regarding the proper fusion of motion generation methods (such as incorporating local reactive policies into long-horizon optimizers). We demonstrate the effectiveness of RMPs on both simulation and real robots, including their ability to naturally and efficiently solve complicated collision avoidance problems previously handled by more complex planners.
We present a new dataset, called Falling Things (FAT), for advancing the state-of-the-art in object detection and 3D pose estimation in the context of robotics. By synthetically combining object models and backgrounds of complex composition and high graphical quality, we are able to generate photorealistic images with accurate 3D pose annotations for all objects in all images. Our dataset contains 60k annotated photos of 21 household objects taken from the YCB dataset. For each image, we provide the 3D poses, per-pixel class segmentation, and 2D/3D bounding box coordinates for all objects. To facilitate testing different input modalities, we provide mono and stereo RGB images, along with registered dense depth images. We describe in detail the generation process and statistical analysis of the data.
We present a system to infer and execute a human-readable program from a real-world demonstration. The system consists of a series of neural networks to perform perception, program generation, and program execution. Leveraging convolutional pose machines, the perception network reliably detects the bounding cuboids of objects in real images even when severely occluded, after training only on synthetic images using domain randomization. To increase the applicability of the perception network to new scenarios, the network is formulated to predict in image space rather than in world space. Additional networks detect relationships between objects, generate plans, and determine actions to reproduce a real-world demonstration. The networks are trained entirely in simulation, and the system is tested in the real world on the pick-and-place problem of stacking colored cubes using a Baxter robot.
Learning a policy capable of moving an agent between any two states in the environment is important for many robotics problems involving navigation and manipulation. Due to the sparsity of rewards in such tasks, applying reinforcement learning in these scenarios can be challenging. Common approaches for tackling this problem include reward engineering with auxiliary rewards, requiring domain-specific knowledge or changing the objective. In this work, we introduce a method based on region-growing that allows learning in an environment with any pair of initial and goal states. Our algorithm first learns how to move between nearby states and then increases the difficulty of the start-goal transitions as the agent's performance improves. This approach creates an efficient curriculum for learning the objective behavior of reaching any goal from any initial state. In addition, we describe a method to adaptively adjust expansion of the growing region that allows automatic adjustment of the key exploration hyperparameter to environments with different requirements. We evaluate our approach on a set of simulated navigation and manipulation tasks, where we demonstrate that our algorithm can efficiently learn a policy in the presence of sparse rewards.
We present a system for training deep neural networks for object detection using synthetic images. To handle the variability in real-world data, the system relies upon the technique of domain randomization, in which the parameters of the simulator$-$such as lighting, pose, object textures, etc.$-$are randomized in non-realistic ways to force the neural network to learn the essential features of the object of interest. We explore the importance of these parameters, showing that it is possible to produce a network with compelling performance using only non-artistically-generated synthetic data. With additional fine-tuning on real data, the network yields better performance than using real data alone. This result opens up the possibility of using inexpensive synthetic data for training neural networks while avoiding the need to collect large amounts of hand-annotated real-world data or to generate high-fidelity synthetic worlds$-$both of which remain bottlenecks for many applications. The approach is evaluated on bounding box detection of cars on the KITTI dataset.
We revisit the problem of visual depth estimation in the context of autonomous vehicles. Despite the progress on monocular depth estimation in recent years, we show that the gap between monocular and stereo depth accuracy remains large$-$a particularly relevant result due to the prevalent reliance upon monocular cameras by vehicles that are expected to be self-driving. We argue that the challenges of removing this gap are significant, owing to fundamental limitations of monocular vision. As a result, we focus our efforts on depth estimation by stereo. We propose a novel semi-supervised learning approach to training a deep stereo neural network, along with a novel architecture containing a machine-learned argmax layer and a custom runtime (that will be shared publicly) that enables a smaller version of our stereo DNN to run on an embedded GPU. Competitive results are shown on the KITTI 2015 stereo dataset. We also evaluate the recent progress of stereo algorithms by measuring the impact upon accuracy of various design criteria.