IROS 2021 Papers

Teleoperation platforms often require the user to be situated at a fixed location to both visualize and control the movement of the robot and thus do not provide the operator with much mobility. One example is in existing robotic surgery solutions that require the surgeons to be away from the patient, attached to consoles where their heads must be fixed and their arms can only move in a limited space. This creates a barrier between physicians and patients that does not exist in normal surgery. To address this issue, we propose a mobile telesurgery solution where the surgeons are no longer mechanically limited to control consoles and are able to teleoperate the robots from the patient bedside, using their arms equipped with wireless sensors and viewing the endoscope video via optical see-through head-mounted displays (HMDs). We evaluate the feasibility and efficiency of our user interaction method compared to a standard surgical robotic manipulator via two tasks with different levels of required dexterity. The results indicate that with sufficient training our proposed platform can attain similar efficiency while providing added mobility for the operator.

Mobile robots are increasingly being deployed in public spaces such as shopping malls, airports, and urban sidewalks. Most of these robots are designed with human-aware motion planning capabilities but are not designed to communicate with pedestrians. Pedestrians encounter these robots without prior understanding of the robots’ behaviour, which can cause discomfort, confusion, and delayed social acceptance. In this research, we explore the common human-robot interaction at a doorway or bottleneck in a structured environment. We designed and evaluated communication cues used by a robot when yielding to a pedestrian in this scenario. We conducted an online user study with 102 participants using videos of a set of robot-to-human yielding cues. Results show that a Robot Retreating cue was the most socially acceptable cue. Repeated measures and Friedman’s ANOVAs on components of social acceptability were statistically significant (p = .01) and had small and medium effect sizes (ηp2 = .04, ηp2 = .08). The results of this work help guide the development of mobile robots for public spaces.

In many multi-robot applications, mobile worker robots are often engaged in performing some tasks repetitively by following pre-computed trajectories. As these robots are battery-powered, they need to get recharged at regular intervals. We envision that, in the future, a few mobile recharger robots will be employed to supply charge to the energy-deficient worker robots recurrently to keep the overall efficiency of the system optimized. In this setup, we need to find the time instants and locations for the meeting of the worker robots and recharger robots optimally. We present a Satisfiability Modulo Theory (SMT)-based approach that captures the activities of the robots in the form of constraints in a sufficiently long finite-length time window (hypercycle) whose repetitions provide their perpetual behavior. Our SMT encoding ensures that for a chosen length of the hypercycle, the total waiting time of the worker robots due to charge constraints is minimized under certain condition, and close to optimal when the condition does not hold. Moreover, the recharger robots follow the most energy-efficient trajectories. We show the efficacy of our approach by comparing it with another variant of the SMT-based method which is not scalable but provides an optimal solution globally, and with a greedy algorithm.

This paper presents a novel strategy for the autonomous deployment of Micro Aerial Vehicle scouts through constricted aperture-like ingress points, by narrowly fitting and launching them with a high-precision Mobile Manipulation robot. A significant problem during exploration and reconnaissance into highly unstructured environments, such as indoor collapsed ones, is the encountering of impassable areas due to their constricted and rigid nature. We propose that a heterogeneous robotic system-of-systems armed with manipulation capabilities while also ferrying a fleet of microsized aerial agents, can deploy the latter through constricted apertures that marginally fit them in size, thus allowing them to act as scouts and resume the reconnaissance mission. This work’s contribution is twofold: first, it proposes active-vision based aperture detection to locate candidate ingress points and a hierarchical search-based aperture profile analysis to position a MAV’s body through them, and secondly it presents and experimentally demonstrates the novelty of a system-of-systems approach which leverages mobile manipulation to deploy other robots which are otherwise incapable of entering through extremely narrow openings.

The system design and algorithm development of mobile 3D printing robots need a realistic simulation. They require a mobile robot simulation platform to interoperate with a physics-based material simulation for handling interactions between the time-variant deformable 3D printing materials and other simulated rigid bodies in the environment, which is not available for roboticists yet. To bridge this gap and enable the real-time simulation of mobile 3D printing processes, we develop a simulation framework that includes particle-based viscoelastic fluid simulation and particle-to-mesh conversion in the widely adopted Gazebo robotics simulator, avoiding the bottlenecks of traditional additive manufacturing simulation approaches. This framework is the first of its kind that enables the simulation of robot arms or mobile manipulators together with viscoelastic fluids. The method is tested using various material properties and multiple collaborating robots to demonstrate its simulation ability for the robots to plan and control the printhead trajectories and to visually sense at the same time the printed fluid materials as a free-form mesh. The scalability as a function of available material particles in the simulation was also studied. A simulation with an average of 5 FPS was achieved on a regular desktop computer.

Despite recent advances in reinforcement learning (RL), its application in safety critical domains like autonomous vehicles is still challenging. Although penalizing RL agents for risky situations can help to learn safe policies, it may also lead to highly conservative behavior. In this paper, we propose a distributional RL framework in order to learn adaptive policies which allow to tune their level of conservativity at run-time based on the desired comfort and utility. Using a proactive safety verification approach, the proposed framework can guarantee that actions generated from RL are failsafe according to the worst-case assumptions. Concurrently, the policy is encouraged to minimize safety interference and generate more comfortable behavior. We trained and evaluated the proposed approach and baseline policies using a high level simulator with a variety of randomized scenarios including several corner cases which rarely happen in reality but are very crucial. In light of our experiments, the behavior of policies learned using distributional RL is adaptive at run-time and robust to the environment uncertainty. Quantitatively, the learned distributional RL agent reduces the average driving time more than 50% compared to the normal DQN policy. It also requires 83% less safety interference compared to the rule-based policy while only slightly increasing the average driving time. We also study sensitivity of the learned policy in environments with higher perception noise and show that our algorithm learns policies that can still drive reliable when the perception noise is two times higher than in the training configuration in automated merging and crossing at occluded intersections.

In the growing elderly population globally, patients with severe movement disorders account for a large proportion. Moreover, the development of intelligent service equipment can better assist them in their daily. This paper proposes a new service robot control system. The brain-computer interface (BCI) based on Steady-State Visual Evoked Potentials (SSVEP) is used to acquire and process electroencephalogram(EEG) signals and output various control commands accordingly. Then, considering the visual fatigue of SSVEP-BCI, we added an object detection method based on Yolov3-tiny and saliency prediction to identify the patient’s selection intention intelligently. The results show that the subject can successfully complete the object delivery task with an average accuracy of 90.3%. The proposed control system can help the patients control a service robot in a more intelligent and friendly way to realize some daily tasks.

As robotic technologies advance and robots move out of factories and labs into the real world, grip on a variety of surfaces (e.g. smooth or rough) in a variety of conditions (e.g. dry or wet) becomes increasingly important. Bioinspired "microspines" have been previously explored, but primarily for vertical climbing applications or for small-scale robots applying low forces (less than 1 N). Further, these works primarily focused on rough surfaces. To advance this area of research, we present a composite material comprising high- friction rubber and compliant nitinol microspines which can passively retract below the surface of the rubber. We show that the composite can support large loads (greater than 75 N) with a high coefficient of friction on both smooth and rough surfaces (p > 1.1), outperforming microspines alone on smooth surfaces and rubber alone on rough surfaces, especially when wet and oily. Further, due to the retraction of the microspines, the composite does not damage relatively soft, smooth surfaces, like wood flooring. We also test durability, and show that it is improved by microspine compliance, and test the effects of varying microspine diameter, angle, and tip shape. Finally, we demonstrate that a small RC car can climb steeper slopes and stop more quickly in wet conditions with microspines.

Wearable IMU sensing systems have widely been used in the study of human motion. For example, gait analysis using wearable inertial sensor systems is a tool used by clinicians to discriminate between typical and pathological walking. Similarly, key descriptors can be identified in spontaneous kicking to distinguish between typical and atypical motor development in infants. Oftentimes in human applications, precise placement of inertial sensors is difficult due to the irregular shape of human limbs. Without precise placement and alignment of the inertial sensors, meaningful joint kinematic data are difficult to extract as the orientation of the joint axes are unknown in the sensor's local frame. So, for applications where precise alignment may not be possible, a necessary first step is to identify the joint axes with respect to the local frame.In this work, we propose a method for the identification of joint axes for multiple degree of freedom (multi-DOF) joints in a kinematic chain using acceleration and angular rate data. This method couples a thresholding activity detection algorithm with a principal component analysis (PCA) dimensionality reduction technique. Furthermore, this method is validated on mimicked kicking data from a NAO robot. This method can determine joint axes of a kinematic chain from simultaneous movement data within an error ratio of 0.09.

This work developed meta-learning control policies to achieve fast online adaptation to different changing conditions, which generate diverse and robust locomotion. The proposed method updates the interaction model constantly, samples feasible sequences of actions of estimated state-action trajectories, and then applies the optimal actions to maximize the reward. To achieve online model adaptation, our proposed method learns different latent vectors of each training condition, which is selected online based on newly collected data from the past 10 samples within 0.2s. Our work designs appropriate state space and reward functions, and optimizes feasible actions in an MPC fashion which are sampled directly in the joint space with constraints, hence requiring no prior design or training of specific gaits. We further demonstrated the robot’s capability of detecting unexpected changes during the interaction and adapting the control policy in less than 0.2s. The extensive validation on the SpotMicro robot in a physics simulation shows adaptive and robust locomotion skills under changing ground friction, external pushes, and different robot dynamics including motor failures and the whole leg amputation.

As multi-robot systems (MRS) are widely used in various tasks such as natural disaster response and social security, people enthusiastically expect an MRS to be ubiquitous that a general user without heavy training can easily operate. However, humans have various preferences on balancing between task performance and safety, imposing different requirements onto MRS control. Failing to comply with preferences makes people feel difficult in operation and decreases human willingness of using an MRS. Therefore, to improve social acceptance as well as performance, there is an urgent need to adjust MRS behaviors according to human preferences before triggering human corrections, which increases cognitive load. In this paper, a novel Meta Preference Learning (MPL) method was developed to enable an MRS to fast adapt to user preferences. MPL based on meta learning mechanism can quickly assess human preferences from limited instructions; then, a neural network based preference model adjusts MRS behaviors for preference adaption. To validate method effectiveness, a task scenario "An MRS searches victims in an earthquake disaster site" was designed; 20 human users were involved to identify preferences as {"aggressive", "medium", "reserved"}; based on user guidance and domain knowledge, about 20,000 preferences were simulated to cover different operations related to {"task quality", "task progress", "robot safety"}. The effectiveness of MPL in preference adaption was validated by the reduced duration and frequency of human interventions.

With the availability of many datasets tailored for autonomous driving in real-world urban scenes, semantic segmentation for urban driving scenes achieves significant progress. However, semantic segmentation for off-road, unstructured environments is not widely studied. Directly applying existing segmentation networks often results in performance degradation as they cannot overcome intrinsic problems in such environments, such as illumination changes. In this paper, a built-in memory module for semantic segmentation is proposed to overcome these problems. The memory module stores significant representations of training images as memory items. In addition to the encoder embedding like items together, the proposed memory module is specifically designed to cluster together instances of the same class even when there are significant variances in embedded features. Therefore, it makes segmentation networks better deal with unexpected illumination changes. A triplet loss is used in training to minimize redundancy in storing discriminative representations of the memory module. The proposed memory module is general so that it can be adopted in a variety of networks. We conduct experiments on the Robot Unstructured Ground Driving (RUGD) dataset and RELLIS dataset, which are collected from off-road, unstructured natural environments. Experimental results show that the proposed memory module improves the performance of existing segmentation networks and contributes to capturing unclear objects over various off-road, unstructured natural scenes with equivalent computational cost and network parameters. As the proposed method can be integrated into compact networks, it presents a viable approach for resource-limited small autonomous platforms.

A promising characteristic of Deep Reinforcement Learning (DRL) is its capability to learn optimal policy in an end-to-end manner without relying on feature engineering. However, most approaches assume a fully observable state space, i.e. fully observable Markov Decision Processes (MDPs). In real-world robotics, this assumption is unpractical, because of issues such as sensor sensitivity limitations and sensor noise, and the lack of knowledge about whether the observation design is complete or not. These scenarios lead to Partially Observable MDPs (POMDPs). In this paper, we propose Long-Short-Term-Memory-based Twin Delayed Deep Deterministic Policy Gradient (LSTM-TD3) by introducing a memory component to TD3, and compare its performance with other DRL algorithms in both MDPs and POMDPs. Our results demonstrate the significant advantages of the memory component in addressing POMDPs, including the ability to handle missing and noisy observation data.

Finding an occluded object in a lateral access environment such as a shelf or cabinet is a problem that arises in many contexts such as warehouses, retail, healthcare, shipping, and homes. While this problem, known as mechanical search, is well-studied in overhead access environments, lateral access environments introduce constraints on the poses of objects and on available grasp actions, and pushing actions are preferred to preserve the environment structure. We propose LAX-RAY (Lateral Access maXimal Reduction in support Area of occupancY distribution): a system that combines target object occupancy distribution prediction with a mechanical search policy that sequentially pushes occluding objects to reveal a given target object. For scenarios with extruded polygonal objects, we introduce two lateral-access search policies that encode a history of predicted target distributions and can plan up to three actions into the future. We introduce a First-Order Shelf Simulator (FOSS) and use it to evaluate these policies in 800 simulated random shelf environments per policy. We also evaluate in 5 physical shelf environments using a Fetch robot with an embedded PrimeSense RGBD Camera and an attached pushing blade. The policies outperform baselines by up to 25% in simulation and up to 60% in physical experiments. Additionally, the two-step prediction policy is the highest performing in simulation for 8 objects with a 69% success rate, suggesting a tradeoff between future information and prediction errors. Code, videos, and supplementary material can be found at https://sites.google.com/berkeley.edu/lax-ray.

The combination of a dexterous continuum robot and magnetic resonance imaging can potentially improve surgical precision and minimize brain manipulation in a minimally invasive neurosurgical procedure. In this work, a seven degree-of-freedom (DoF) continuum neurosurgical robot was developed. The main innovation lies in the design of a safe and robust switching mechanism and gear-based quick-connect mechanism that, respectively, allow selective actuation of the 6-DoF end effector using only three motors and highly efficient end effector exchange. Its performance has been validated in experiments involving multi-segment dexterous motion. We also evaluated the robotic system on a human cadaver head in a clinical 3-Tesla MRI. The entire workflow of robotic system set-up was implemented, confirming its clinical feasibility. The signal-to-noise ratio (SNR) drop was consistently less than 6% throughout various stages of end effector motion.

The increasing complexity of modern robotic systems and the environments they operate in necessitates the formal consideration of safety in the presence of imperfect measurements. In this paper we propose a rigorous framework for safety-critical control of systems with erroneous state estimates. We develop this framework by leveraging Control Barrier Functions (CBFs) and unifying the method of Backup Sets for synthesizing control invariant sets with robustness requirements—the end result is the synthesis of Measurement-Robust Control Barrier Functions (MR-CBFs). This provides theoretical guarantees on safe behavior in the presence of imperfect measurements and improved robustness over standard CBF approaches. We demonstrate the efficacy of this framework both in simulation and experimentally on a Segway platform using an onboard stereo-vision camera for state estimation.

We propose a decentralized approach that simultaneously allocates and decomposes high level tasks among various robots. The approach exploits HTN structures and algorithms, that are used within an auction-based allocation scheme, and aims at dealing with complex tasks with causal or temporal relations. The paper formalizes the approach, and depicts how HTN planning processes are used to estimate bids and distribute tasks. Results on a statistical series of coverage problems are presented and their performance is assessed through a comparison with a state of the art algorithm.

This paper proposes a multi-objective visionbased navigation network for biomimetic underwater robots to cope with scientific observation, target selection, and obstacle avoidance in marine missions. Structurally, a stereo block attention module is first constructed to serially extract the channel and spatial attention portion of the real-time visual feedback. Next, the parallax attention mechanism is introduced to enable the network to excavate implicit parallax information in stereo pairs, effectively eliminating the oscillation of the network output in the presence of ambiguous visual input. Further, with the assistance of other low-cost sensors, the proposed navigation network can be expanded in some largescale application scenarios, such as sparse coral observation. Finally, underwater simulations reveal that the proposed method obtains significantly improved control effect and real-time ability, compared with other related works. In particular, based on a self-developed biomimetic robotic dolphin, collision-free simulations with a cumulative distance beyond 1000 m were carried out and validated the effectiveness and the superiority of the navigation network, where both dense and sparse targets were fully tested. The robotic dolphin can not only successfully conduct accurate coral observation without collision, but also quest the observation targets as much as possible in the area where the observation targets are concentrated. The proposed network provides an intelligent and efficient navigation scheme for autonomous underwater operation of small-size underwater robots.

In this paper, we propose a novel Deep Reinforcement Learning approach to address the mapless navigation problem, in which the locomotion actions of a humanoid robot are taken online based on the knowledge encoded in learned models. Planning happens by generating open-loop trajectories in a learned latent space that captures the dynamics of the environment. Our planner considers visual (RGB images) and non-visual observations (e.g., attitude estimations). This confers the agent upon awareness not only of the scenario, but also of its own state. In addition, we incorporate a termination likelihood predictor model as an auxiliary loss function of the control policy, which enables the agent to anticipate terminal states of success and failure. In this manner, the sample efficiency of the approach for episodic tasks is increased. Our model is evaluated on the NimbRo-OP2X humanoid robot that navigates in scenes avoiding collisions efficiently in simulation and with the real hardware.