Sim2Real transfer, particularly for manipulation policies relying on RGB images, remains a critical challenge in robotics due to the significant domain shift between syn-thetic and real-world visual data. In this paper, we propose SplatSim, a novel framework that leverages Gaussian Splatting as the primary rendering primitive to reduce the Sim2Real gap for RGB-based manipulation policies. By replacing traditional mesh representations with Gaussian Splats in simulators, SplatSim produces highly photorealistic synthetic data while maintaining the scalability and cost-efficiency of simulation. We demonstrate the effectiveness of our framework by training manipulation policies within SplatSim and deploying them in the real world in a zero-shot manner, achieving an average success rate of 86.25%, compared to 97.5% for policies trained on real-world data. Videos can be found on our project page: https://splatsim.github.io

Offline reinforcement learning can enable policy learning from pre-collected, sub-optimal datasets without online interactions. This makes it ideal for real-world robots and safety-critical scenarios, where collecting online data or expert demonstrations is slow, costly, and risky. However, most existing offline RL works assume the dataset is already labeled with the task rewards, a process that often requires significant human effort, especially when ground-truth states are hard to ascertain (e.g., in the real-world). In this paper, we build on prior work, specifically RL-VLM-F, and propose a novel system that automatically generates reward labels for offline datasets using preference feedback from a vision-language model and a text description of the task. Our method then learns a policy using offline RL with the reward-labeled dataset. We demonstrate the system’s applicability to a complex real-world robot-assisted dressing task, where we first learn a reward function using a vision-language model on a sub-optimal offline dataset, and then we use the learned reward to employ Implicit Q learning to develop an effective dressing policy. Our method also performs well in simulation tasks involving the manipulation of rigid and deformable objects, and significantly outperforms baselines such as behavior cloning and inverse RL. In summary, we propose a new system that enables automatic reward labeling and policy learning from unlabeled, sub-optimal offline datasets. Videos can be found on our project website1.

Many robot manipulation tasks can be framed as geometric reasoning tasks, where an agent must be able to precisely manipulate an object into a position that satisfies the task from a set of initial conditions. Often, task success is defined based on the relationship between two objects - for instance, hanging a mug on a rack. In such cases, the solution should be equivariant to the initial position of the objects as well as the agent, and invariant to the pose of the camera. This poses a challenge for learning systems which attempt to solve this task by learning directly from high-dimensional demonstrations: the agent must learn to be both equivariant as well as precise, which can be challenging without any inductive biases about the problem. In this work, we propose a method for precise relative pose prediction which is provably SE(3)-equivariant, can be learned from only a few demonstrations, and can generalize across variations in a class of objects. We accomplish this by factoring the problem into learning an SE(3) invariant task-specific representation of the scene and then interpreting this representation with novel geometric reasoning layers which are provably SE(3) equivariant. We demonstrate that our method can yield substantially more precise placement predictions in simulated placement tasks than previous methods trained with the same amount of data, and can accurately represent relative placement relationships data collected from real-world demonstrations. Supplementary information and videos can be found at https://sites.google.com/view/reldist-iclr-2023.

Reward engineering has long been a challenge in Reinforcement Learning (RL) research, as it often requires extensive human effort and iterative processes of trial-and-error to design effective reward functions. In this paper, we propose RL-VLM-F, a method that automatically generates reward functions for agents to learn new tasks, using only a text description of the task goal and the agent's visual observations, by leveraging feedbacks from vision language foundation models (VLMs). The key to our approach is to query these models to give preferences over pairs of the agent's image observations based on the text description of the task goal, and then learn a reward function from the preference labels, rather than directly prompting these models to output a raw reward score, which can be noisy and inconsistent. We demonstrate that RL-VLM-F successfully produces effective rewards and policies across various domains — including classic control, as well as manipulation of rigid, articulated, and deformable objects — without the need for human supervision, outperforming prior methods that use large pretrained models for reward generation under the same assumptions. Videos can be found on our project website: https://rlvlmf2024.github.io/

We present RoboGen, a generative robotic agent that automatically learns diverse robotic skills at scale via generative simulation. RoboGen leverages the latest advancements in foundation and generative models. Instead of directly adapting these models to produce policies or low-level actions, we advocate for a generative scheme, which uses these models to automatically generate diversified tasks, scenes, and training supervisions, thereby scaling up robotic skill learning with minimal human supervision. Our approach equips a robotic agent with a self-guided propose-generate-learn cycle: the agent first proposes interesting tasks and skills to develop, and then generates simulation environments by populating pertinent assets with proper spatial configurations. Afterwards, the agent decomposes the proposed task into sub-tasks, selects the optimal learning approach (reinforcement learning, motion planning, or trajectory optimization), generates required training supervision, and then learns policies to acquire the proposed skill. Our fully generative pipeline can be queried repeatedly, producing an endless stream of skill demonstrations associated with diverse tasks and environments.

Relative placement tasks are an important category of tasks in which one object needs to be placed in a desired pose relative to another object. Previous work has shown success in learning relative placement tasks from just a small number of demonstrations when using relational reasoning networks with geometric inductive biases. However, such methods cannot flexibly represent multimodal tasks, like a mug hanging on any of n racks. We propose a method that incorporates additional properties that enable learning multimodal relative placement solutions, while retaining the provably translation-invariant and relational properties of prior work. We show that our method is able to learn precise relative placement tasks with only 10-20 multimodal demonstrations with no human annotations across a diverse set of objects within a category. Supplementary information can be found on the website: https://sites.google.com/view/tax-posed/home.

Safe Reinforcement Learning (RL) plays an important role in applying RL algorithms to safety-critical real-world applications, addressing the trade-off between maximizing rewards and adhering to safety constraints. This work introduces a novel approach that combines RL with trajectory optimization to manage this trade-off effectively. Our approach embeds safety constraints within the action space of a modified Markov Decision Process (MDP). The RL agent produces a sequence of actions that are transformed into safe trajectories by a trajectory optimizer, thereby effectively ensuring safety and increasing training stability. This novel approach excels in its performance on challenging Safety Gym tasks, achieving significantly higher rewards and near-zero safety violations during inference. The method’s real-world applicability is demonstrated through a safe and effective deployment in a real robot task of box-pushing around obstacles. Further insights are available from the videos and appendix on our website: https://sites.google.com/view/safemdp.

Autonomous systems that efficiently utilize tools can assist humans in completing many common tasks such as cooking and cleaning. However, current systems fall short of matching human-level of intelligence in terms of adapting to novel tools. Prior works based on affordance often make strong assumptions about the environments and cannot scale to more complex, contact-rich tasks. In this work, we tackle this challenge and explore how agents can learn to use previously unseen tools to manipulate deformable objects. We propose to learn a generative model of the tool-use trajectories as a sequence of tool point clouds, which generalizes to different tool shapes. Given any novel tool, we first generate a tool-use trajectory and then optimize the sequence of tool poses to align with the generated trajectory. We train a single model on four different challenging deformable object manipulation tasks, using demonstration data from only one tool per task. The model generalizes to various novel tools, significantly outperforming baselines. We further test our trained policy in the real world with unseen tools, where it achieves the performance comparable to human. Additional materials can be found on our project website.1

This paper introduces DiffTORI, which utilizes $\textbf{Diff}$erentiable $\textbf{T}$rajectory $\textbf{O}$ptimization as the policy representation to generate actions for deep $\textbf{R}$einforcement and $\textbf{I}$mitation learning. Trajectory optimization is a powerful and widely used algorithm in control, parameterized by a cost and a dynamics function. The key to our approach is to leverage the recent progress in differentiable trajectory optimization, which enables computing the gradients of the loss with respect to the parameters of trajectory optimization. As a result, the cost and dynamics functions of trajectory optimization can be learned end-to-end. DiffTORI addresses the “objective mismatch” issue of prior model-based RL algorithms, as the dynamics model in DiffTORI is learned to directly maximize task performance by differentiating the policy gradient loss through the trajectory optimization process. We further benchmark DiffTORI for imitation learning on standard robotic manipulation task suites with high-dimensional sensory observations and compare our method to feedforward policy classes as well as Energy-Based Models (EBM) and Diffusion. Across 15 model based RL tasks and 35 imitation learning tasks with high-dimensional image and point cloud inputs, DiffTORI outperforms prior state-of-the-art methods in both domains.

Predicting how the world can evolve in the future is crucial for motion planning in autonomous systems. Classical methods are limited because they rely on costly human annotations in the form of semantic class labels, bounding boxes, and tracks or HD maps of cities to plan their motion - and thus are difficult to scale to large unlabeled datasets. One promising self-supervised task is 3D point cloud forecasting [11, 18–20] from unannotated LiDAR sequences. We show that this task requires algorithms to implicitly capture (1) sensor extrinsics (i.e., the egomotion of the autonomous vehicle), (2) sensor intrinsics (i.e., the sampling pattern specific to the particular LiDAR sensor), and (3) the shape and motion of other objects in the scene. But autonomous systems should make predictions about the world and not their sensors! To this end, we factor out (1) and (2) by recasting the task as one of spacetime (4D) occupancy forecasting. But because it is expensive to obtain ground-truth 4D occupancy, we “render” point cloud data from 4D occupancy predictions given sensor extrinsics and intrinsics, allowing one to train and test occupancy algorithms with unannotated LiDAR sequences. This also allows one to evaluate and compare point cloud forecasting algorithms across diverse datasets, sensors, and vehicles.

Thin plastic bags are ubiquitous in retail stores, healthcare, food handling, recycling, homes, and school lunchrooms. They are challenging both for perception (due to specularities and occlusions) and for manipulation (due to the dynamics of their 3D deformable structure). We formulate the task of “bagging:” manipulating common plastic shopping bags with two handles from an unstructured initial state to an open state where at least one solid object can be inserted into the bag and lifted for transport. We propose a self-supervised learning framework where a dual-arm robot learns to recognize the handles and rim of plastic bags using UV-fluorescent markings; at execution time, the robot does not use UV markings or UV light. We propose the AutoBag algorithm, where the robot uses the learned perception model to open a plastic bag through iterative manipulation. We present novel metrics to evaluate the quality of a bag state and new motion primitives for reorienting and opening bags based on visual observations. In physical experiments, a YuMi robot using AutoBag is able to open bags and achieve a success rate of 16/30 for inserting at least one item across a variety of initial bag configurations. Supplementary material is available at https://sites.google.com/view/autobag.

State estimation is one of the greatest challenges for cloth manipulation due to cloth's high dimensionality and self-occlusion. Prior works propose to identify the full state of crumpled clothes by training a mesh reconstruction model in simulation. However, such models are prone to suffer from a sim-to-real gap due to differences between cloth simulation and the real world. In this work, we propose a self-supervised method to finetune a mesh reconstruction model in the real world. Since the full mesh of crumpled cloth is difficult to obtain in the real world, we design a special data collection scheme and an action-conditioned model-based cloth tracking method to generate pseudo-labels for self-supervised learning. By finetuning the pretrained mesh reconstruction model on this pseudo-labeled dataset, we show that we can improve the quality of the reconstructed mesh without requiring human annotations, and improve the performance of downstream manipulation task. More visualizations and results can be found on our project website.

We study the problem of learning graph dynamics of deformable objects that generalizes to unknown physical properties. Our key insight is to leverage a latent representation of elastic physical properties of cloth-like deformable objects that can be extracted, for example, from a pulling interaction. In this paper we propose EDO-Net (Elastic Deformable Object - Net), a model of graph dynamics trained on a large variety of samples with different elastic properties that does not rely on ground-truth labels of the properties. EDO-Net jointly learns an adaptation module, and a forward-dynamics module. The former is responsible for extracting a latent representation of the physical properties of the object, while the latter leverages the latent representation to predict future states of cloth-like objects represented as graphs. We evaluate EDO-Net both in simulation and real world, assessing its capabilities of: 1) generalizing to unknown physical properties, 2) transferring the learned representation to new downstream tasks.

Physical interaction with textiles, such as assistive dressing or household tasks, requires advanced dexterous skills. The complexity of textile behavior during stretching and pulling is influenced by the material properties of the yarn and by the textile's construction technique, which are often unknown in real-world settings. Moreover, identification of physical properties of textiles through sensing commonly available on robotic platforms remains an open problem. To address this, we introduce Elastic Context (EC), a method to encode the elasticity of textiles using stress-strain curves adapted from textile engineering for robotic applications. We employ EC to learn generalized elastic behaviors of textiles and examine the effect of EC dimension on accurate force modeling of real-world non-linear elastic behaviors.

We formulate grasp learning as a neural field and present Neural Grasp Distance Fields (NGDF). Here, the input is a 6D pose of a robot end effector and output is a distance to a continuous manifold of valid grasps for an object. In contrast to current approaches that predict a set of discrete candidate grasps, the distance-based NGDF representation is easily interpreted as a cost, and minimizing this cost produces a successful grasp pose. This grasp distance cost can be incorporated directly into a trajectory optimizer for joint optimization with other costs such as trajectory smoothness and collision avoidance. During optimization, as the various costs are balanced and minimized, the grasp target is allowed to smoothly vary, as the learned grasp field is continuous. We evaluate NGDF on joint grasp and motion planning in simulation and the real world, outperforming baselines by 63 % execution success while generalizing to unseen query poses and unseen object shapes. Project page: https://sites.google.com/view/neural-grasp-distance-fields.

Many fabric handling and 2D deformable material tasks in homes and industries require singulating layers of material such as opening a bag or arranging garments for sewing. In contrast to methods requiring specialized sensing or end effectors, we use only visual observations with ordinary parallel jaw grippers. We propose SLIP: Singulating Layers using Interactive Perception, and apply SLIP to the task of autonomous bagging. We develop SLIP-Bagging, a bagging algorithm that manipulates a plastic or fabric bag from an unstructured state and uses SLIP to grasp the top layer of the bag to open it for object insertion. In physical experiments, a YuMi robot achieves a success rate of 67% to 81% across bags of a variety of materials, shapes, and sizes, significantly improving in success rate and generality over prior work. Experiments also suggest that SLIP can be applied to tasks such as singulating layers of folded cloth and garments. Supplementary material is available at https://sites.google.com/view/slip-bagging/.

Liquid state estimation is important for robotics tasks such as pouring; however, estimating the state of transparent liquids is a challenging problem. We propose a novel segmentation pipeline that can segment transparent liquids such as water from a static, RGB image without requiring any manual annotations or heating of the liquid for training. Instead, we use a generative model that is capable of translating images of colored liquids into synthetically generated transparent liquid images, trained only on an unpaired dataset of colored and transparent liquid images. Segmentation labels of colored liquids are obtained automatically using background subtraction. Our experiments show that we are able to accurately predict a segmentation mask for transparent liquids without requiring any manual annotations. We demonstrate the utility of transparent liquid segmentation in a robotic pouring task that controls pouring by perceiving the liquid height in a transparent cup. Accompanying video and supplementary materials can be found at https://sites.google.com/view/transparentliquidpouring.

Robotic manipulation of cloth has applications ranging from fabrics manufacturing to handling blankets and laundry. Cloth manipulation is challenging for robots largely due to their high degrees of freedom, complex dynamics, and severe self-occlusions when in folded or crumpled configurations. Prior work on robotic manipulation of cloth relies primarily on vision sensors alone, which may pose challenges for fine-grained manipulation tasks such as grasping a desired number of cloth layers from a stack of cloth. In this paper, we propose to use tactile sensing for cloth manipulation; we attach a tactile sensor (ReSkin) to one of the two fingertips of a Franka robot and train a classifier to determine whether the robot is grasping a specific number of cloth layers. During test-time experiments, the robot uses this classifier as part of its policy to grasp one or two cloth layers using tactile feedback to determine suitable grasping points. Experimental results over 180 physical trials suggest that the proposed method outperforms baselines that do not use tactile feedback and has better generalization to unseen cloth compared to methods that use image classifiers. Code, data, and videos are available at https://sites.google.com/view/reskin-cloth.

3D object detection plays an important role in autonomous driving and other robotics applications. However, these detectors usually require training on large amounts of annotated data that is expensive and time-consuming to collect. Instead, we propose leveraging large amounts of unlabeled point cloud videos by semi-supervised learning of 3D object detectors via temporal graph neural networks. Our insight is that temporal smoothing can create more accurate detection results on unlabeled data, and these smoothed detections can then be used to retrain the detector. We learn to perform this temporal reasoning with a graph neural network, where edges represent the relationship between candidate detections in different time frames. After semi-supervised learning, our method achieves state-of-the-art detection performance on the challenging nuScenes [3] and H3D [19] benchmarks, compared to baselines trained on the same amount of labeled data. Project and code are released at https://www.jianrenw.com/SOD-TGNN/.

Safe local motion planning for autonomous driving in dynamic environments requires forecasting how the scene evolves. Practical autonomy stacks adopt a semantic object-centric representation of a dynamic scene and build object detection, tracking, and prediction modules to solve forecasting. However, training these modules comes at an enormous human cost of manually annotated objects across frames. In this work, we explore future freespace as an alternative representation to support motion planning. Our key intuition is that it is important to avoid straying into occupied space regardless of what is occupying it. Importantly, computing ground-truth future freespace is annotation-free. First, we explore freespace forecasting as a self-supervised learning task. We then demonstrate how to use forecasted freespace to identify collision-prone plans from off-the-shelf motion planners. Finally, we propose future freespace as an additional source of annotation-free supervision. We demonstrate how to integrate such supervision into the learning-based planners. Experimental results on nuScenes and CARLA suggest both approaches lead to a significant reduction in collision rates.1

Active sensing through the use of Adaptive Depth Sensors is a nascent field, with potential in areas such as Advanced driver-assistance systems (ADAS). They do however require dynamically driving a laser / light-source to a specific location to capture information, with one such class of sensor being the Triangulation Light Curtains (LC). In this work, we introduce a novel approach that exploits prior depth distributions from RGB cameras to drive a Light Curtain’s laser line to regions of uncertainty to get new measurements. These measurements are utilized such that depth uncertainty is reduced and errors get corrected recursively. We show real-world experiments that validate our approach in outdoor and driving settings, and demonstrate qualitative and quantitative improvements in depth RMSE when RGB cameras are used in tandem with a Light Curtain.

Pose estimation is a basic module in many robot manipulation pipelines. Estimating the pose of objects in the environment can be useful for grasping, motion planning, or manipulation. However, current state-of-the-art methods for pose estimation either rely on large annotated training sets or simulated data. Further, the long training times for these methods prohibit quick interaction with novel objects. To address these issues, we introduce a novel method for zero-shot object pose estimation in clutter. Our approach uses a hypothesis generation and scoring framework, with a focus on learning a scoring function that generalizes to objects not used for training. We achieve zero-shot generalization by rating hypotheses as a function of unordered point differences. We evaluate our method on challenging datasets with both textured and untextured objects in cluttered scenes and demonstrate that our method significantly outperforms previous methods on this task. We also demonstrate how our system can be used by quickly scanning and building a model of a novel object, which can immediately be used by our method for pose estimation. Our work allows users to estimate the pose of novel objects without requiring any retraining. Additional information can be found on our website https://bokorn.github.io/zephyr/

Visual data in autonomous driving perception, such as camera image and LiDAR point cloud, can be interpreted as a mixture of two aspects: semantic feature and geometric structure. Semantics come from the appearance and context of objects to the sensor, while geometric structure is the actual 3D shape of point clouds. Most detectors on LiDAR point clouds focus only on analyzing the geometric structure of objects in real 3D space. Unlike previous works, we propose to learn both semantic feature and geometric structure via a unified multi-view framework. Our method exploits the nature of LiDAR scans - 2D range images, and applies well-studied 2D convolutions to extract semantic features. By fusing semantic and geometric features, our method outperforms state-of-the-art approaches in all categories by a large margin. The methodology of combining semantic and geometric features provides a unique perspective of looking at the problems in real-world 3D point cloud detection.

Recent advances in 3D sensing have created unique challenges for computer vision. One fundamental challenge is finding a good representation for 3D sensor data. Most popular representations (such as PointNet) are proposed in the context of processing truly 3D data (e.g. points sampled from mesh models), ignoring the fact that 3D sensored data such as a LiDAR sweep is in fact 2.5D. We argue that representing 2.5D data as collections of (x,y,z) points fundamentally destroys hidden information about freespace. In this paper, we demonstrate such knowledge can be efficiently recovered through 3D raycasting and readily incorporated into batch-based gradient learning. We describe a simple approach to augmenting voxel-based networks with visibility: we add a voxelized visibility map as an additional input stream. In addition, we show that visibility can be combined with two crucial modifications common to state-of-the-art 3D detectors: synthetic data augmentation of virtual objects and temporal aggregation of LiDAR sweeps over multiple time frames. On the NuScenes 3D detection benchmark, we show that, by adding an additional stream for visibility input, we can significantly improve the overall detection accuracy of a state-of-the-art 3D detector.

When interacting with highly dynamic environments, scene flow allows autonomous systems to reason about the non-rigid motion of multiple independent objects. This is of particular interest in the field of autonomous driving, in which many cars, people, bicycles, and other objects need to be accurately tracked. Current state-of-the-art methods require annotated scene flow data from autonomous driving scenes to train scene flow networks with supervised learning. As an alternative, we present a method of training scene flow that uses two self-supervised losses, based on nearest neighbors and cycle consistency. These self-supervised losses allow us to train our method on large unlabeled autonomous driving datasets; the resulting method matches current state-of-the-art supervised performance using no real world annotations and exceeds state-of-the-art performance when combining our self-supervised approach with supervised learning on a smaller labeled dataset.

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/.

Cloth detection and manipulation is a common task in domestic and industrial settings, yet such tasks remain a challenge for robots due to cloth deformability. Furthermore, in many cloth-related tasks like laundry folding and bed making, it is crucial to manipulate specific regions like edges and corners, as opposed to folds. In this work, we focus on the problem of segmenting and grasping these key regions. Our approach trains a network to segment the edges and corners of a cloth from a depth image, distinguishing such regions from wrinkles or folds. We also provide a novel algorithm for estimating the grasp location, direction, and directional uncertainty from the segmentation. We demonstrate our method on a real robot system and show that it outperforms baseline methods on grasping success. Video and other supplementary materials are available at: https://sites.google.com/view/cloth-segmentation.

3D object trackers usually require training on large amounts of annotated data that is expensive and time-consuming to collect. Instead, we propose leveraging vast unlabeled datasets by self-supervised metric learning of 3D object trackers, with a focus on data association. Large scale annotations for unlabeled data are cheaply obtained by automatic object detection and association across frames. We show how these self-supervised annotations can be used in a principled manner to learn point-cloud embeddings that are effective for 3D tracking. We estimate and incorporate uncertainty in self-supervised tracking to learn more robust embeddings, without needing any labeled data. We design embeddings to differentiate objects across frames, and learn them using uncertainty-aware self-supervised training. Finally, we demonstrate their ability to perform accurate data association across frames, towards effective and accurate 3D tracking. Project videos and code are at https://jianrenw.github.io/Self-Supervised-3D-Data-Association/.

3D multi-object tracking (MOT) is an essential component for many applications such as autonomous driving and assistive robotics. Recent work on 3D MOT focuses on developing accurate systems giving less attention to practical considerations such as computational cost and system complexity. In contrast, this work proposes a simple real-time 3D MOT system. Our system first obtains 3D detections from a LiDAR point cloud. Then, a straightforward combination of a 3D Kalman filter and the Hungarian algorithm is used for state estimation and data association. Additionally, 3D MOT datasets such as KITTI evaluate MOT methods in the 2D space and standardized 3D MOT evaluation tools are missing for a fair comparison of 3D MOT methods. Therefore, we propose a new 3D MOT evaluation tool along with three new metrics to comprehensively evaluate 3D MOT methods. We show that, although our system employs a combination of classical MOT modules, we achieve state-of-the-art 3D MOT performance on two 3D MOT benchmarks (KITTI and nuScenes). Surprisingly, although our system does not use any 2D data as inputs, we achieve competitive performance on the KITTI 2D MOT leaderboard. Our proposed system runs at a rate of 207.4 FPS on the KITTI dataset, achieving the fastest speed among all modern MOT systems. To encourage standardized 3D MOT evaluation, our code is publicly available at http://www.xinshuoweng.com/projects/AB3DMOT.

Robots that are trained to perform a task in a fixed environment often fail when facing unexpected changes to the environment due to a lack of exploration. We propose a principled way to adapt the policy for better exploration in changing sparse-reward environments. Unlike previous works which explicitly model environmental changes, we analyze the relationship between the value function and the optimal exploration for a Gaussian-parameterized policy and show that our theory leads to an effective strategy for adjusting the variance of the policy, enabling fast adapt to changes in a variety of sparse-reward environments.

Reinforcement learning is known to be sample inefficient, preventing its application to many real-world problems, especially with high dimensional observations like images. Transferring knowledge from other auxiliary tasks is a powerful tool for improving the learning efficiency. However, the usage of auxiliary tasks has been limited so far due to the difficulty in selecting and combining different auxiliary tasks. In this work, we propose a principled online learning algorithm that dynamically combines different auxiliary tasks to speed up training for reinforcement learning. Our method is based on the idea that auxiliary tasks should provide gradient directions that, in the long term, help to decrease the loss of the main task. We show in various environments that our algorithm can effectively combine a variety of different auxiliary tasks and achieves significant speedup compared to previous heuristic approches of adapting auxiliary task weights.

Shape completion, the problem of estimating the complete geometry of objects from partial observations, lies at the core of many vision and robotics applications. In this work, we propose Point Completion Network (PCN), a novel learning-based approach for shape completion. Unlike existing shape completion methods, PCN directly operates on raw point clouds without any structural assumption (e.g. symmetry) or annotation (e.g. semantic class) about the underlying shape. It features a decoder design that enables the generation of fine-grained completions while maintaining a small number of parameters. Our experiments show that PCN produces dense, complete point clouds with realistic structures in the missing regions on inputs with various levels of incompleteness and noise, including cars from LiDAR scans in the KITTI dataset.

Reinforcement learning (RL) is a powerful technique to train an agent to perform a task; however, an agent that is trained using RL is only capable of achieving the single task that is specified via its reward function. Such an approach does not scale well to settings in which an agent needs to perform a diverse set of tasks, such as navigating to varying positions in a room or moving objects to varying locations. Instead, we propose a method that allows an agent to automatically discover the range of tasks that it is capable of performing in its environment. We use a generator network to propose tasks for the agent to try to accomplish, each task being specified as reaching a certain parametrized subset of the state-space. The generator network is optimized using adversarial training to produce tasks that are always at the appropriate level of difficulty for the agent, thus automatically producing a curriculum. We show that, by using this framework, an agent can efficiently and automatically learn to perform a wide set of tasks without requiring any prior knowledge of its environment, even when only sparse rewards are available. Videos and code available at https://sites.google.com/view/goalgeneration4rl.

For many applications of reinforcement learning it can be more convenient to specify both a reward function and constraints, rather than trying to design behavior through the reward function. For example, systems that physically interact with or around humans should satisfy safety constraints. Recent advances in policy search algorithms (Mnih et al., 2016, Schulman et al., 2015, Lillicrap et al., 2016, Levine et al., 2016) have enabled new capabilities in high-dimensional control, but do not consider the constrained setting. We propose Constrained Policy Optimization (CPO), the first general-purpose policy search algorithm for constrained reinforcement learning with guarantees for near-constraint satisfaction at each iteration. Our method allows us to train neural network policies for high-dimensional control while making guarantees about policy behavior all throughout training. Our guarantees are based on a new theoretical result, which is of independent interest: we prove a bound relating the expected returns of two policies to an average divergence between them. We demonstrate the effectiveness of our approach on simulated robot locomotion tasks where the agent must satisfy constraints motivated by safety.

Although learning-based methods have great potential for robotics, one concern is that a robot that updates its parameters might cause large amounts of damage before it learns the optimal policy. We formalize the idea of safe learning in a probabilistic sense by defining an optimization problem: we desire to maximize the expected return while keeping the expected damage below a given safety limit. We study this optimization for the case of a robot manipulator with safety-based torque limits. We would like to ensure that the damage constraint is maintained at every step of the optimization and not just at convergence. To achieve this aim, we introduce a novel method which predicts how modifying the torque limit, as well as how updating the policy parameters, might affect the robot's safety. We show through a number of experiments that our approach allows the robot to improve its performance while ensuring that the expected damage constraint is not violated during the learning process.

Non-prehensile manipulation, such as pushing, is an important function for robots to move objects and is sometimes preferred as an alternative to grasping. However, due to unknown frictional forces, pushing has been proven a difficult task for robots. We explore the use of reinforcement learning to train a robot to robustly push an object. In order to deal with the sample complexity of training such a method, we train the pushing policy in simulation and then transfer this policy to the real world. In order to ease the transfer from simulation, we propose to use modularity to separate the learned policy from the raw inputs and outputs; rather than training “end-to-end,” we decompose our system into modules and train only a subset of these modules in simulation. We further demonstrate that we can incorporate prior knowledge about the task into the state space and the reward function to speed up convergence. Finally, we introduce “reward guiding” to modify the reward function and further reduce the training time. We demonstrate, in both simulation and real-world experiments, that such an approach can be used to reliably push an object from many initial positions and orientations. Videos available at https://goo.gl/B7LtY3.

Some robots must repeatedly interact with a fixed set of objects in their environment. To operate correctly, it is helpful for the robot to be able to recognize the object instances that it repeatedly encounters. However, current methods for recognizing object instances require that, during training, many pictures are taken of each object from a large number of viewing angles. This procedure is slow and requires much manual effort before the robot can begin to operate in a new environment. We have developed a novel procedure for training a neural network to recognize a set of objects from just a single training image per object. To obtain robustness to changes in viewpoint, we take advantage of a supplementary dataset in which we observe a separate (non-overlapping) set of objects from multiple viewpoints. After pre-training the network in a novel multi-stage fashion, the network can robustly recognize new object instances given just a single training image of each object. If more images of each object are available, the performance improves. We perform a thorough analysis comparing our novel training procedure to traditional neural network pre-training techniques as well as previous state-of-the-art approaches including keypoint-matching, template-matching, and sparse coding, and we demonstrate that our method significantly outperforms these previous approaches. Our method can thus be used to easily teach a robot to recognize a novel set of object instances from unknown viewpoints.

Precision tracking is important for predicting the behavior of other cars in autonomous driving. We present a novel method to combine laser and camera data to achieve accurate velocity estimates of moving vehicles. We combine sparse laser points with a high-resolution camera image to obtain a dense colored point cloud. We use a color-augmented search algorithm to align the dense color point clouds from successive time frames for a moving vehicle, thereby obtaining a precise estimate of the tracked vehicle's velocity. Using this alignment method, we obtain velocity estimates at a much higher accuracy than previous methods. Through pre-filtering, we are able to achieve near real time results. We also present an online method for real-time use with accuracies close to that of the full method. We present a novel approach to quantitatively evaluate our velocity estimates by tracking a parked car in a local reference frame in which it appears to be moving relative to the ego vehicle. We use this evaluation method to automatically quantitatively evaluate our tracking performance on 466 separate tracked vehicles. Our method obtains a mean absolute velocity error of 0.27 m/s and an RMS error of 0.47 m/s on this test set. We can also qualitatively evaluate our method by building color 3D car models from moving vehicles. We have thus demonstrated that our method can be used for precision car tracking with applications to autonomous driving and behavior modeling.

Detecting cars in real-world images is an important task for autonomous driving, yet it remains unsolved. The system described in this paper takes advantage of context and scale to build a monocular single-frame image-based car detector that significantly outperforms the baseline. The system uses a probabilistic model to combine multiple forms of evidence for both context and scale to locate cars in a real-world image. We also use scale filtering to speed up our algorithm by a factor of 3.3 compared to the baseline. By using a calibrated camera and localization on a road map, we are able to obtain context and scale information from a single image without the use of a 3D laser. The system outperforms the baseline by an absolute 9.4% in overall average precision and 11.7% in average precision for cars smaller than 50 pixels in height, for which context and scale cues are especially important.

A number of robotic studies have recently turned to biological inspiration in designing control schemes for flexible robots. Examples of such robots include continuous manipulators inspired by the octopus arm. However, the control strategies used by an octopus in moving its arms are still not fully understood. Starting from a dynamic model of an octopus arm and a given set of muscle activations, we develop a simulation technique to characterize the stiffness throughout a motion and at multiple points along the arm. By applying this technique to reaching and bending motions, we gain a number of insights that can help a control engineer design a biologically inspired impedance control scheme for a flexible robot arm. The framework developed is a general one that can be applied to any motion for any dynamic model. We also propose a theoretical analysis to efficiently estimate the stiffness analytically given a set of muscle activations. This analysis can be used to quickly evaluate the stiffness for new static configurations and dynamic movements.

In order to achieve autonomous operation of a vehicle in urban situations with unpredictable traffic, several realtime systems must interoperate, including environment perception, localization, planning, and control. In addition, a robust vehicle platform with appropriate sensors, computational hardware, networking, and software infrastructure is essential. We previously published an overview of Junior, Stanford's entry in the 2007 DARPA Urban Challenge. This race was a closed-course competition which, while historic and inciting much progress in the field, was not fully representative of the situations that exist in the real world. In this paper, we present a summary of our recent research towards the goal of enabling safe and robust autonomous operation in more realistic situations. First, a trio of unsupervised algorithms automatically calibrates our 64-beam rotating LIDAR with accuracy superior to tedious hand measurements. We then generate high-resolution maps of the environment which are subsequently used for online localization with centimeter accuracy. Improved perception and recognition algorithms now enable Junior to track and classify obstacles as cyclists, pedestrians, and vehicles; traffic lights are detected as well. A new planning system uses this incoming data to generate thousands of candidate trajectories per second, choosing the optimal path dynamically. The improved controller continuously selects throttle, brake, and steering actuations that maximize comfort and minimize trajectory error. All of these algorithms work in sun or rain and during the day or night. With these systems operating together, Junior has successfully logged hundreds of miles of autonomous operation in a variety of real-life conditions.