ICRA 2021 Papers

Making decisions in complex driving environments is a challenging task for autonomous agents. Imitation learning methods have great potentials for achieving such a goal. Adversarial Inverse Reinforcement Learning (AIRL) is one of the state-of-art imitation learning methods that can learn both a behavioral policy and a reward function simultaneously, yet it is only demonstrated in simple and static environments where no interactions are introduced. In this paper, we improve and stabilize AIRL’s performance by augmenting it with semantic rewards in the learning framework. Additionally, we adapt the augmented AIRL to a more practical and challenging decision-making task in a highly interactive environment in autonomous driving. The proposed method is compared with four baselines and evaluated by four performance metrics. Simulation results show that the augmented AIRL outperforms all the baseline methods, and its performance is comparable with that of the experts on all of the four metrics.

Safe and efficient navigation through human crowds is an essential capability for mobile robots. Previous work on robot crowd navigation assumes that the dynamics of all agents are known and well-defined. In addition, the performance of previous methods deteriorates in partially observable environments and environments with dense crowds. To tackle these problems, we propose decentralized structural-Recurrent Neural Network (DS-RNN), a novel network that reasons about spatial and temporal relationships for robot decision making in crowd navigation. We train our network with model-free deep reinforcement learning without any expert supervision. We demonstrate that our model outperforms previous methods in challenging crowd navigation scenarios. We successfully transfer the policy learned in the simulator to a real-world TurtleBot 2i.

In this paper, we propose two decentralized approximate algorithms for nested Gaussian processes in multi-robot systems. The distributed implementation is achieved with iterative and consensus methods that facilitate local computations at the expense of inter-robot communications. Moreover, we propose a covariance-based nearest neighbor robot selection strategy that enables a subset of agents to perform predictions. In addition, both algorithms are proved to be consistent. Empirical evaluations with real data illustrate the efficiency of the proposed algorithms.

Connectivity maintenance is crucial for the real world deployment of multi-robot systems, as it ultimately allows the robots to communicate, coordinate and perform tasks in a collaborative way. A connectivity maintenance controller must keep the multi-robot system connected independently from the system’s mission and in the presence of undesired real world effects such as communication delays, model errors, and computational time delays, among others. In this paper we present the implementation, on a real robotic setup, of a connectivity maintenance control strategy based on Control Barrier Functions. During experimentation, we found that the presence of communication delays has a significant impact on the performance of the controlled system, with respect to the ideal case. We propose a heuristic to counteract the effects of communication delays, and we verify its efficacy both in simulation and with physical robot experiments.

In this paper, the circle formation control problem is addressed for a group of cooperative underactuated fish-like robots involving unknown nonlinear dynamics and disturbances. Based on the reinforcement learning and cognitive consistency theory, we propose a decentralized controller without the knowledge of the dynamics of the fish-like robots. The proposed controller can be transferred from simulation to reality. It is only trained in our established simulation environment, and the trained controller can be deployed to real robots without any manual tuning. Simulation results confirm that the proposed model-free robust formation control method is scalable with respect to the group size of the robots and outperforms other representative RL algorithms. Several experiments in the real world verify the effectiveness of our RL-based approach for circle formation control.

Situation awareness is of great importance for autonomous ships. One key aspect is to estimate the sea state in a real-time manner. Considering the ship as a large wave buoy, the sea state can be estimated from motion responses without extra sensors installed. However, it is difficult to associate waves with ship motion through an explicit model since the hydrodynamic effect is hard to model. In this paper, a data-driven model is developed to estimate the sea state based on ship motion data. The ship motion response is analyzed through statistical, temporal, spectral, and wavelet analysis. Features from multi-domain are constructed and an ensemble machine learning model is established. Real-world data is collected from a research vessel operating on the west coast of Norway. Through the validation with the real-world data, the model shows promising performance in terms of significant wave height and peak period.

Laparoscopic Field of View (FOV) control is one of the most fundamental and important components in Minimally Invasive Surgery (MIS), nevertheless the traditional manual holding paradigm may easily bring fatigue to surgical assistants, and misunderstanding between surgeons also hinders assistants to provide a high-quality FOV. Targeting this problem, we here present a data-driven framework to realize an automated laparoscopic optimal FOV control. To achieve this goal, we offline learn a motion strategy of laparoscope relative to the surgeon’s hand-held surgical tool from our in-house surgical videos, developing our control domain knowledge and an optimal view generator. To adjust the laparoscope online, we first adopt a learning-based method to segment the two-dimensional (2D) position of the surgical tool, and further leverage this outcome to obtain its scale-aware depth from dense depth estimation results calculated by our novel unsupervised RoboDepth model only with the monocular camera feedback, hence in return fusing the above real-time 3D position into our control loop. To eliminate the misorientation of FOV caused by Remote Center of Motion (RCM) constraints when moving the laparoscope, we propose a novel rotation constraint using an affine map to minimize the visual warping problem, and a null-space controller is also embedded into the framework to optimize all types of errors in a unified and decoupled manner. Experiments are conducted using Universal Robot (UR) and Karl Storz Laparoscope/Instruments, which prove the feasibility of our domain knowledge and learning enabled framework for automated camera control.

Even though artificial muscles have gained popularity due to their compliant, flexible and compact properties, there currently does not exist an easy way of making informed decisions on the appropriate actuation strategy when designing a muscle-powered robot; thus limiting the transition of such technologies into broader applications. What’s more, when a new muscle actuation technology is developed, it is difficult to compare it against existing robot muscles. To accelerate the development of artificial muscle applications, we propose a data-driven approach for robot muscle actuator selection using Support Vector Machines (SVM). This first-of-its-kind method gives users insight into which actuators fit their specific needs and actuation performance criteria, making it possible for researchers and engineers with little to no prior knowledge of artificial muscles to focus on application design. It also provides a platform to benchmark existing, new or yet-to-be discovered artificial muscle technologies. We test our method on unseen existing robot muscle designs to prove its usability on real-world applications. We provide an open-access, web-searchable interface for easy access to our models that will additionally allow for the continuous contribution of new actuator data from groups around the world to enhance and expand these models.

This paper presents a data-based approach to control robotic systems with partially-observed feedback. First, an open-loop optimization problem is solved to generate the nominal trajectory and then a linear time-varying Autoregressive–Moving-Average (ARMA) model of the system is calculated along the trajectory from the output measurement data. The system is then described in information state, which contains input-output information of the past few steps. Finally, a feedback gain which is calculated by solving a specific LQG problem along the nominal trajectory. The separate design of the open-loop and the closed-loop problem is used following the Decoupled Data-based Control (D2C) approach. Simulation results are also shown for complex models with fluid-structure interaction in the presence of both process and measurement noise.

Multi-robot task allocation is a ubiquitous problem in robotics due to its applicability in a variety of scenarios. Adaptive task-allocation algorithms account for unknown disturbances and unpredicted phenomena in the environment where robots are deployed to execute tasks. However, this adaptivity typically comes at the cost of requiring precise knowledge of robot models in order to evaluate the allocation effectiveness and to adjust the task assignment online. As such, environmental disturbances can significantly degrade the accuracy of the models which in turn negatively affects the quality of the task allocation. In this paper, we leverage Gaussian processes, differential inclusions, and robust control barrier functions to learn environmental disturbances in order to guarantee robust task execution. We show the implementation and the effectiveness of the proposed framework on a real multi-robot system.

In visual robot self-localization, graph-based scene representation and matching have recently attracted research interest as robust and discriminative methods for self-localization. Although effective, their computational and storage costs do not scale well to large-size environments. To alleviate this problem, we formulate self-localization as a graph classification problem and attempt to use the graph convolutional neural network (GCN) as a graph classification engine. A straightforward approach is to use visual feature descriptors that are employed by state-of-the-art self-localization systems, directly as graph node features. However, their superior performance in the original self-localization system may not necessarily be replicated in GCN-based self-localization. To address this issue, we introduce a novel teacher-to-student knowledge-transfer scheme based on rank matching, in which the reciprocal-rank vector output by an off-the-shelf state-of-the-art teacher self-localization model is used as the dark knowledge to transfer. Experiments indicate that the proposed graph-convolutional self-localization network (GCLN) can significantly outperform state-of-the-art self-localization systems, as well as the teacher classifier. The code and dataset are available at https://github.com/KojiTakeda00/Reciprocal_rank_KT_GCN.

We present a novel Deep Reinforcement Learning (DRL) based policy to compute dynamically feasible and spatially aware velocities for a robot navigating among mobile obstacles. Our approach combines the benefits of the Dynamic Window Approach (DWA) in terms of satisfying the robot’s dynamics constraints with state-of-the-art DRL-based navigation methods that can handle moving obstacles and pedestrians well. Our formulation achieves these goals by embedding the environmental obstacles’ motions in a novel low-dimensional observation space. It also uses a novel reward function to positively reinforce velocities that move the robot away from the obstacle’s heading direction leading to significantly lower number of collisions. We evaluate our method in realistic 3-D simulated environments and on a real differential drive robot in challenging dense indoor scenarios with several walking pedestrians. We compare our method with state-of-the-art collision avoidance methods and observe significant improvements in terms of success rate (up to 33% increase), number of dynamics constraint violations (up to 61% decrease), and smoothness. We also conduct ablation studies to highlight the advantages of our observation space formulation, and reward structure.

We present DRACO, a method for Dense Reconstruction And Canonicalization of Object shape from one or more RGB images. Canonical shape reconstruction— estimating 3D object shape in a coordinate space canonicalized for scale, rotation, and translation parameters—is an emerging paradigm that holds promise for a multitude of robotic applications. Prior approaches either rely on painstakingly gathered dense 3D supervision, or produce only sparse canonical representations, limiting real-world applicability. DRACO performs dense canonicalization using only weak supervision in the form of camera poses and semantic keypoints at train time. During inference, DRACO predicts dense object-centric depth maps in a canonical coordinate-space, solely using one or more RGB images of an object. Extensive experiments on canonical shape reconstruction and pose estimation show that DRACO is competitive or superior to fully-supervised methods.

In this paper we present DOT (Dynamic Object Tracking), a front-end that added to existing SLAM systems can significantly improve their robustness and accuracy in highly dynamic environments. DOT combines instance segmentation and multi-view geometry to generate masks for dynamic objects in order to allow SLAM systems based on rigid scene models to avoid such image areas in their optimizations.To determine which objects are actually moving, DOT segments first instances of potentially dynamic objects and then, with the estimated camera motion, tracks such objects by minimizing the photometric reprojection error. This short-term tracking improves the accuracy of the segmentation with respect to other approaches. In the end, only actually dynamicmasks are generated.We have evaluated DOT with ORB-SLAM 2 [1] in three public datasets. Our results show that our approach improves significantly the accuracy and robustness of ORB-SLAM2, especially in highly dynamic scenes.

We propose a Deep Interaction Prediction Network (DIPN) for learning to predict complex interactions that ensue as a robot end-effector pushes multiple objects, whose physical properties, including size, shape, mass, and friction coefficients may be unknown a priori. DIPN "imagines" the effect of a push action and generates an accurate synthetic image of the predicted outcome. DIPN is shown to be sample efficient when trained in simulation or with a real robotic system. The high accuracy of DIPN allows direct integration with a grasp network, yielding a robotic manipulation system capable of executing challenging clutter removal tasks while being trained in a fully self-supervised manner. The overall network demonstrates intelligent behavior in selecting proper actions between push and grasp for completing clutter removal tasks and significantly outperforms the previous state-of-the-art. Remarkably, DIPN achieves even better performance on the real robotic hardware system than in simulation.

Exploration in environments with sparse rewards remains a challenging problem in Deep Reinforcement Learning (DRL). For the off-policy method, it usually needs a large number of training samples. With the growing dimensions of state and action space, this method becomes more and more sample-inefficient. In this paper, we propose a novel fast exploration method for off-policy reinforcement learning, called Density-based Intrinsic Motivation and Self-adaptive Action Noise (DIMSAN). Our main contribution is twofold: (1) We propose a Density-based Intrinsic Motivation (DIM) method. It introduces a new intrinsic-reward generation mechanism based on samples’ density estimation during experience replay and encourages the agent to seek novel and unfamiliar states. (2) We propose a Self-adaptive Action Noise (SAN) to deal with the exploration-exploitation tradeoffs, which could automatically change the exploration step through adding adaptive action space noise. The synergy between DIM and SAN could guide the agent to search the state and action space with high efficiency. We evaluate our method on the benchmark manipulation tasks and the designed challenging ones. Empirical results show that our method outperforms the existing methods in terms of convergence speed and sample efficiency, especially in challenging tasks.

This paper presents a contact-aided inertial-kinematic floating base estimation for humanoid robots considering an evolution of the state and observations over matrix Lie groups. This is achieved through the application of a geometrically meaningful estimator which is characterized by concentrated Gaussian distributions. The configuration of a floating base system like a humanoid robot usually requires the knowledge of an additional six degrees of freedom which describes its base position-and-orientation. This quantity usually cannot be measured and needs to be estimated. A matrix Lie group, encapsulating the position-and-orientation and linear velocity of the base link, feet positions-and-orientations and Inertial Measurement Units’ biases, is used to represent the state while relative positions-and-orientations of contact feet from forward kinematics are used as observations. The proposed estimator exhibits fast convergence for large initialization errors owing to choice of uncertainty parametrization. An experimental validation is done on the iCub humanoid platform.

Telesurgery can be hindered by high-latency and low-bandwidth communication networks, often found in austere settings. Even delays of less than one second are known to negatively impact surgeries. To tackle the effects of connectivity associated with telerobotic surgeries, we propose the DESERTS framework. DESERTS provides a novel simulator interface where the surgeon can operate directly on a virtualized reality simulation and the activities are mirrored in a remote robot, almost simultaneously. Thus, the surgeon can perform the surgery uninterrupted, while high-level commands are extracted from his motions and are sent to a remote robotic agent. The simulated setup mirrors the remote environment, including an alpha-blended view of the remote scene. The framework abstracts the actions into atomic surgical maneuvers (surgemes) which eliminate the need to transmit compressed video information. This system uses a deep learning based architecture to perform live recognition of the surgemes executed by the operator. The robot then executes the received surgemes, thereby achieving semi-autonomy. The framework’s performance was tested on a peg transfer task. We evaluated the accuracy of the recognition and execution module independently as well as during live execution. Furthermore, we assessed the framework’s performance in the presence of increasing delays. Notably, the system maintained a task success rate of 87% from no-delays to 5 seconds of delay.

An Adaptive Cruise Control (ACC) system allows vehicles to maintain a desired headway distance to a preceding vehicle automatically. It is increasingly adopted by commercial vehicles. Recent research demonstrates that the effective use of ACC can improve the traffic flow through the adaptation of the headway distance in response to the current traffic conditions. In this paper, we demonstrate that a state-of-the- art intelligent ACC system performs poorly on highways with ramps due to the limitation of the model-based approaches that do not take into account appropriately the traffic dynamics on ramps in determining the optimal headway distance. We then propose a dynamic adaptive cruise control system (D- ACC) based on deep reinforcement learning that adapts the headway distance effectively according to dynamically changing traffic conditions for both the main road and ramp to optimize the traffic flow. Extensive simulations are performed with a combination of a traffic simulator (SUMO) and vehicle-to- everything communication (V2X) network simulator (Veins) under numerous traffic scenarios. We demonstrate that D-ACC improves the traffic flow by up to 70% compared with a state- of-the-art intelligent ACC system in a highway segment with a ramp.