ICRA 2023 Papers

Preference-based reinforcement learning (PbRL) can enable robots to learn to perform tasks based on an individual's preferences without requiring a hand-crafted re-ward function. However, existing approaches either assume access to a high-fidelity simulator or analytic model or take a model-free approach that requires extensive, possibly unsafe online environment interactions. In this paper, we study the benefits and challenges of using a learned dynamics model when performing PbRL. In particular, we provide evidence that a learned dynamics model offers the following benefits when performing PbRL: (1) preference elicitation and policy optimization require significantly fewer environment interactions than model-free PbRL, (2) diverse preference queries can be synthesized safely and efficiently as a byproduct of standard model-based RL, and (3) reward pre-training based on suboptimal demonstrations can be performed without any environmental interaction. Our paper provides empirical ev-idence that learned dynamics models enable robots to learn customized policies based on user preferences in ways that are safer and more sample efficient than prior preference learning approaches. Supplementary materials and code are available at https://sites.google.com/berkeley.edu/mop-rl.

Efficient multi-robot task allocation (MRTA) is fundamental to various time-sensitive applications such as disaster response, warehouse operations, and construction. This paper tackles a particular class of these problems that we call MRTA-collective transport or MRTA-CT - here tasks present varying workloads and deadlines, and robots are subject to flight range, communication range, and payload constraints. For large instances of these problems involving 100s-1000's of tasks and 10s-100s of robots, traditional non-learning solvers are often time-inefficient, and emerging learning-based policies do not scale well to larger-sized problems without costly retraining. To address this gap, we use a recently proposed encoder-decoder graph neural network involving Capsule networks and multi-head attention mechanism, and innovatively add topological descriptors (TD) as new features to improve transferability to unseen problems of similar and larger size. Persistent homology is used to derive the TD, and proximal policy optimization is used to train our TD-augmented graph neural network. The resulting policy model compares favorably to state-of-the-art non-learning baselines while being much faster. The benefit of using TD is readily evident when scaling to test problems of size larger than those used in training.

State estimation is an essential part of autonomous systems. Integrating the Ultra-Wideband (UWB) technique has been shown to correct the long-term estimation drift and bypass the complexity of loop closure detection. However, few works on robotics treat UWB as a stand-alone state estimation solution. The primary purpose of this work is to investigate planar pose estimation using only UWB range measurements. We prove the excellent property of a two-step scheme, which says we can refine a consistent estimator to be asymptotically efficient by one step of Gauss-Newton iteration. Grounded on this result, we design the GN-ULS estimator, which reduces the computation time significantly compared to previous methods and presents the possibility of using only UWB for real-time state estimation.

We propose an algorithm for solving the time-dependent shortest path problem in flow fields where the FIFO (first-in-first-out) assumption is violated. This problem variant is important for autonomous vehicles in the ocean, for example, that cannot arbitrarily hover in a fixed position and that are strongly influenced by time-varying ocean currents. Although polynomial-time solutions are available for discrete-time problems, the continuous-time non-FIFO case is NP-hard with no known relevant special cases. Our main result is to show that this problem can be solved in polynomial time if the edge travel time functions are piecewise-constant, agreeing with existing worst-case bounds for FIFO problems with restricted slopes. We present a minimum-time algorithm for graphs that allows for paths with finite-length cycles, and then embed this algorithm within an asymptotically optimal sampling-based framework to find time-optimal paths in flows. The algorithm relies on an efficient data structure to represent and manipulate piecewise-constant functions and is straightforward to implement. We illustrate the behaviour of the algorithm in an example based on a common ocean vortex model.

Data-driven learning based methods have recently been particularly successful at learning robust locomotion controllers for a variety of unstructured terrains. Prior work has shown that incorporating good locomotion priors in the form of trajectory generators (TGs) is effective at efficiently learning complex locomotion skills. However, defining a good, single TG as tasks/environments become increasingly more complex remains a challenging problem as it requires extensive tuning and risks reducing the effectiveness of the prior. In this paper, we present Evolved Environmental Trajectory Generators (EETG), a method that learns a diverse set of specialised locomotion priors using Quality-Diversity algorithms while maintaining a single policy within the Policies Modulating TG (PMTG) architecture. The results demonstrate that EETG enables a quadruped robot to successfully traverse a wide range of environments, such as slopes, stairs, rough terrain, and balance beams. Our experiments show that learning a diverse set of specialized TG priors is significantly (5 times) more efficient than using a single, fixed prior when dealing with a wide range of environments.

Real-world robotic manipulation tasks remain an elusive challenge, since they involve both fine-grained environment interaction, as well as the ability to plan for long-horizon goals. Although deep reinforcement learning (RL) methods have shown encouraging results when planning end-to-end in high-dimensional environments, they remain fundamentally limited by poor sample efficiency due to inefficient exploration, and by the complexity of credit assignment over long horizons. In this work, we present Efficient Learning of High-Level Plans from Play (ELF-P), a framework for robotic learning that bridges motion planning and deep RL to achieve long-horizon complex manipulation tasks. We leverage task-agnostic play data to learn a discrete behavioral prior over object-centric primitives, modeling their feasibility given the current context. We then design a high-level goal-conditioned policy which (1) uses primitives as building blocks to scaffold complex long-horizon tasks and (2) leverages the behavioral prior to accelerate learning. We demonstrate that ELF-P has significantly better sample efficiency than relevant baselines over multiple realistic manipulation tasks and learns policies that can be easily transferred to physical hardware.

Robotic deployments in human environments have motivated the need for autonomous systems to be able to interact with humans and solve tasks effectively. Human demonstrations of tasks can be used to infer underlying task specifications, commonly modeled with temporal logic. State-of-the-art methods have developed Bayesian inference tools to estimate a temporal logic formula from a sequence of demon-strations. The current work proposes the use of experiment design to choose environments for humans to perform these demonstrations. This reduces the number of demonstrations needed to estimate the unknown ground truth formula with low error. A novel computationally efficient strategy is proposed to generate informative environments by using an optimal planner as the model for the demonstrator. Instead of evaluating all possible environments, the search space reduces to the placement of informative orderings of likely eventual goals along an optimal planner's solution. A human study with 600 demonstrations from 20 participants for 4 tasks on a 2D interface validates the proposed hypothesis and empirical performance benefit in terms of convergence and error over baselines. The human study dataset is also publicly shared.

Modeling scene geometry using implicit neural representation has revealed its advantages in accuracy, flexibility, and low memory usage. Previous approaches have demonstrated impressive results using color or depth images but still have difficulty handling poor light conditions and large-scale scenes. Methods taking global point cloud as input require accurate registration and ground truth coordinate labels, which limits their application scenarios. In this paper, we propose a new method that uses sparse LiDAR point clouds and rough odometry to reconstruct fine-grained implicit occupancy field efficiently within a few minutes. We introduce a new loss function that supervises directly in 3D space without 2D rendering, avoiding information loss. We also manage to refine poses of input frames in an end-to-end manner, creating consistent geometry without global point cloud registration. As far as we know, our method is the first to reconstruct implicit scene representation from LiDAR-only input. Experiments on synthetic and real-world datasets, including indoor and outdoor scenes, prove that our method is effective, efficient, and accurate, obtaining comparable results with existing methods using dense input.

Bundle adjustment (BA) is a well-studied fundamental problem in the robotics and vision community. In man-made environments, coplanar points and lines are ubiquitous. However, the number of works on bundle adjustment with coplanar points and lines is relatively small. This paper focuses on this special BA problem, referred to as \pi-\mathbf{BA}\pi-\mathbf{BA}. For a point or a line on a plane, we derive a new constraint to describe the relationship among two poses and the plane, called \pi\pi-constraint. We distribute \pi\pi-constraints into different groups. Each group is called a \pi\pi-factor. We prove that, with some simple preprocessing, the computational complexity associated with a \pi\pi-factor in the Levenberg-Marquardt (LM) algorithm is O(1)O(1), independent of the number of \pi\pi-constraints packed into the \pi\pi-factor. In \pi-\mathbf{BA}, \pi\pi-\mathbf{BA}, \pi-factors replace original reprojection errors. One problem is how to divide \pi\pi-constraints into \pi\pi-factors. Different strategies may result in different numbers of \pi\pi-factors, which in turn affects the efficiency. It is difficult to get the optimal division. We present a greedy algorithm to overcome this problem. Experimental results verify that our algorithm can significantly accelerate the computation.

Bimanual manipulation is important for building intelligent robots that unlock richer skills than single arms. We consider a multi-object bimanual rearrangement task, where a reinforcement learning (RL) agent aims to jointly control two arms to rearrange these objects as fast as possible. Solving this task efficiently is challenging for an RL agent due to the requirement of discovering precise intra-arm coordination in an exponentially large control space. We develop a symmetry-aware actor-critic framework that leverages the interchangeable roles of the two manipulators in the bimanual control setting to reduce the policy search space. To handle the compositionality over multiple objects, we augment training data with an object-centric relabeling technique. The overall approach produces an RL policy that can rearrange up to 8 objects with a success rate of over 70% in simulation. We deploy the policy to two Franka Panda arms and further show a successful demo on human-robot collaboration. Videos can be found at https://sites.google.com/view/bimanual.

Real-time knowledge of the vehicle mass is valuable for several applications, mainly: active safety systems design and energy consumption optimization. This work describes a novel strategy for mass estimation in static and dynamic conditions. First, when the vehicle is powered-up, an initial estimation is given by observing the variations of one suspension deflection sensor mounted on the rear. Then, the estimation is refined based on conditioned and filtered longitudinal and lateral motions. In this study, we suggest using these extracted events on two different algorithms, namely: the recursive least squares and the prior-recursive Bayesian inference. That is to express the results in a deterministic and statistical sense. Both simulations and experimental tests show that our approach encompasses the benefits of various works in the literature, preeminently, robustness to resistive loads, fast convergence, and minimal instrumentation.

We have developed biped robots with a passive dynamic walking mechanism. This study proposes a compass model with a wobbling mass connected to the upper body and oscillating in the horizontal direction to clarify the influence of the horizontal dynamics of the upper body on bipedal walking. The limit cycles of the model were numerically searched, and their stability and energy efficiency was investigated. Several qualitatively different limit cycles were obtained depending mainly on the spring constant that supports the wobbling mass. Specific types of solutions decreased the stability while reducing the risk of accidental falling and improving the energy efficiency. The obtained results were attributed to the wobbling mass moving in the opposite direction to the upper body, thereby preventing large changes in acceleration and deceleration while walking. The relationship between the locomotion of the proposed model and the actual biped robot and human gaits was investigated.

Visual odometry is important for plenty of applications such as autonomous vehicles, and robot navigation. It is challenging to conduct visual odometry in textureless scenes or environments with sudden illumination changes where popular feature-based methods or direct methods cannot work well. To address this challenge, some edge-based methods have been proposed, but they usually struggle between the efficiency and accuracy. In this work, we propose a novel visual odometry approach called EdgeVO, which is accurate, efficient, and robust. By efficiently selecting a small set of edges with certain strategies, we significantly improve the computational efficiency without sacrificing the accuracy. Compared to existing edge-based method, our method can significantly reduce the computational complexity while maintaining similar accuracy or even achieving better accuracy. This is attributed to that our method removes useless or noisy edges. Experimental results on the TUM datasets indicate that EdgeVO significantly outperforms other methods in terms of efficiency, accuracy and robustness.

The insufficient number of annotated thermal infrared (TIR) image datasets not only hinders TIR image-based deep learning networks to have comparable performances to that of RGB but it also limits the supervised learning of TIR image-based tasks with challenging labels. As a remedy, we propose a modified multidomain RGB to TIR image translation model focused on edge preservation to employ annotated RGB images with challenging labels. Our proposed method not only preserves key details in the original image but also leverages the optimal TIR style code to portray accurate TIR characteristics in the translated image, when applied on both synthetic and real world RGB images. Using our translation model, we have enabled the supervised learning of deep TIR image-based optical flow estimation and object detection that ameliorated in deep TIR optical flow estimation by reduction in end point error by 56.5% on average and the best object detection mAP of 23.9% respectively. Our code and supplementary materials are available at https://github.com/rpmsnu/sRGB-TIR.

Given point cloud input, the problem of 6-DoF grasp pose detection is to identify a set of hand poses in SE(3) from which an object can be successfully grasped. This important problem has many practical applications. Here we propose a novel method and neural network model that enables better grasp success rates relative to what is available in the literature. The method takes standard point cloud data as input and works well with single-view point clouds observed from arbitrary viewing directions. Videos and code are available at https://haojhuang.github.io/edge_grasp_page/.

Current robotic hand manipulation narrowly operates with objects in predictable positions in limited environments. Thus, when the location of the target object deviates severely from the expected location, a robot sometimes responds in an unexpected way, especially when it operates with a human. For safe robot operation, we propose the EXit-aware Object Tracker (EXOT) on a robot hand camera that recognizes an object's absence during manipulation. The robot decides whether to proceed by examining the tracker's bounding box output containing the target object. We adopt an out-of-distribution classifier for more accurate object recognition since trackers can mistrack a background as a target object. To the best of our knowledge, our method is the first approach of applying an out-of-distribution classification technique to a tracker output. We evaluate our method on the first-person video benchmark dataset, TREK-150, and on the custom dataset, RMOT-223, that we collect from the UR5e robot. Then we test our tracker on the UR5e robot in real-time with a conveyor-belt sushi task, to examine the tracker's ability to track target dishes and to determine the exit status. Our tracker shows 38% higher exit-aware performance than a baseline method. The dataset and the code will be released at https://github.com/hskAlena/EXOT.

We present EWareNet, a novel intent and affect-aware social robot navigation algorithm among pedestrians. Our approach predicts the trajectory-based pedestrian intent from gait sequence, which is then used for intent-guided navigation taking into account social and proxemic constraints. We propose a transformer-based model that works on commodity RGB-D cameras mounted onto a moving robot. Our intent prediction routine is integrated into a mapless navigation scheme and makes no assumptions about the environment of pedestrian motion. Our navigation scheme consists of a novel obstacle profile representation methodology that is dynamically adjusted based on the pedestrian pose, intent, and affect. The navigation scheme is based on a reinforcement learning algorithm that takes pedestrian intent and robot's impact on pedestrian intent into consideration, in addition to the environmental configuration. We outperform current state-of-art algorithms for intent prediction from 3D gaits.

Among various sensors for assisted and autonomous driving systems, automotive radar has been considered as a robust and low-cost solution even in adverse weather or lighting conditions. With the recent development of radar technologies and open-sourced annotated data sets, semantic segmentation with radar signals has become very promising. However, existing methods are either computationally expensive or discard significant amounts of valuable information from raw 3D radar signals by reducing them to 2D planes via averaging. In this work, we introduce ERASE-Net, an Efficient RAdar SEgmentation Network to segment the raw radar signals semantically. The core of our approach is the novel detect-then-segment method for raw radar signals. It first detects the center point of each object, then extracts a compact radar signal representation, and finally performs semantic segmentation. We show that our method can achieve superior performance on radar semantic segmentation task compared to the state-of-the-art (SOTA) technique. Furthermore, our approach requires up to 20×less computational resources. Finally, we show that the proposed ERASE-Net can be compressed by 40% without significant loss in performance, significantly more than the SOTA network, which makes it a more promising candidate for practical automotive applications.

We propose EMS®, a cloud-enabled massive computational experiment management system supporting high-throughput computational robotics research. Compared to existing systems, EMS® features a sky-based pipeline orchestrator which allows us to exploit heterogeneous computing environments painlessly (e.g., on-premise clusters, public clouds, edge devices) to optimally deploy large-scale computational jobs (e.g., with more than millions of computational hours) in an integrated fashion. Cornerstoned on this sky-based pipeline orchestrator, this paper introduces three abstraction layers of the EMS® software architecture: (i) Configuration management layer focusing on automatically enumerating experimental configurations; (ii) Dependency management layer focusing on managing the complex task dependencies within each experimental configuration; (iii) Computation management layer focusing on optimally executing the computational tasks using the given computing resource. Such an architectural design greatly increases the scalability and reproducibility of data-driven robotics research leading to much-improved productivity. To demonstrate this point, we compare EMS® with more traditional approaches on an offline reinforcement learning problem for training mobile robots. Our results show that EMS® outperforms more traditional approaches in two magnitudes of orders (in terms of experimental high throughput and cost) with only several lines of code change. We also exploit EMS® to develop mobile robot, robot arm, and bipedal applications, demonstrating its applicability to numerous robot applications.