ICRA 2022 Papers

In this paper we investigate the influence interfaces and feedback have on human-robot trust levels when operating in a shared physical space. The task we use is specifying a “no-go” region for a robot in an indoor environment. We evaluate three styles of interface (physical, AR, and map-based) and four feedback mechanisms (no feedback, robot drives around the space, an AR “fence”, and the region marked on the map). Our evaluation looks at both usability and trust. Specifically, if the participant trusts that the robot “knows” where the no-go region is and their confidence in the robot's ability to avoid that region. We use both self-reported and indirect measures of trust and usability. Our key findings are: 1) interfaces and feedback do influence levels of trust; 2) the participants largely preferred a mixed interface-feedback pair, where the modality for the interface differed from the feedback.

Human-Swarm Interaction (HSI) is a fast-growing research area in swarm robotics. One challenging aspect of HSI is facilitating effective handling of the many degrees-of-freedom present in robot swarms by humans. One emergent option is the use of Augmented Reality (AR) systems to encode information. AR based interfaces can help provide human operators with visual cues about the swarm's states and control to facilitate decision-making. In research settings, AR systems can address issues such as limited availability of lab spaces, limited access to robotics resources, and the need for the ability to simulate dynamic environments with which robots and humans can interact. Further, to make swarm robotics more accessible and ubiquitous, HSI systems that support remote interaction would allow humans to interact with robot swarms and multi-robot systems regardless of the geographical distance between humans and swarms. Taking these into consideration, we aim to investigate the effectiveness of AR based interfaces as tools for remote interaction in HSI systems. We developed a simple AR based interface and evaluated its effectiveness against an unaugmented interface, by means of remote human user studies where a human operator would control a team of robots remotely through a video call. Our finding suggests that augmentation can improve control accuracy and reduce collision safety violations when performing navigation tasks. Through experimental surveys, it is shown that operators with varying levels of robotics and technology experience overwhelmingly prefer the augmented interface to facilitate swarm control. These results suggest that AR-based interfaces are effective in improving the control experience in remote HSI.

We believe that the future of robot motion planning will look very different than how it looks today: instead of complex collision avoidance trajectories with a brittle dependence on sensing and estimation of the environment, motion plans should consist of smooth, simple trajectories and be executed by robots that are not afraid of making contact. Here we present a “contact-aware” controller which continues to execute a given trajectory despite unexpected collisions while keeping the contact force stable and small. We introduce a quadratic programming (QP) formulation, which minimizes a trajectory-tracking error subject to quasistatic dynamics and contact-force constraints. Compared with the classical null-space projection technique, the inequality constraint on contact forces in the proposed QP controller allows for more gentle release when the robot comes out of contact. In the quasistatic dynamics model, control actions consist only of commanded joint positions, allowing the QP controller to run on stiffness-controlled robots which do not have a straightforward torque-control interface nor accurate dynamic models. The effectiveness of the proposed QP controller is demonstrated on a KUKA iiwa arm. Project video: https://youtu.be/M-7JMQRkiPk.

This paper introduces an efficient direct visual odometry (VO) algorithm using points and lines. Pixels on lines are generally adopted in direct methods. However, the original photometric error is only defined for points. It seems difficult to extend it to lines. In previous works, the collinear constraints for points on lines are either ignored [1] or introduce heavy computational load into the resulting optimization system [2]. This paper extends the photometric error for lines. We prove that the 3D points of the points on a 2D line are determined by the inverse depths of the endpoints of the 2D line, and derive a closed-form solution for this problem. This property can significantly reduce the number of variables to speed up the optimization, and can make the collinear constraint exactly satisfied. Furthermore, we introduce a two-step method to further accelerate the optimization, and prove the convergence of this method. The experimental results show that our algorithm outperforms the state-of-the-art direct VO algorithms.

Quality-Diversity (QD) algorithms are powerful exploration algorithms that allow robots to discover large repertoires of diverse and high-performing skills. However, QD algorithms are sample inefficient and require millions of evaluations. In this paper, we propose Dynamics-Aware Quality-Diversity (DA-QD), a framework to improve the sample efficiency of QD algorithms through the use of dynamics models. We also show how DA-QD can then be used for continual acquisition of new skill repertoires. To do so, we incrementally train a deep dynamics model from experience obtained when performing skill discovery using QD. We can then perform QD exploration in imagination with an imagined skill repertoire. We evaluate our approach on three robotic experiments. First, our experiments show DA-QD is 20 times more sample efficient than existing QD approaches for skill discovery. Second, we demonstrate learning an entirely new skill repertoire in imagination to perform zero-shot learning. Finally, we show how DA-QD is useful and effective for solving a long horizon navigation task and for damage adaptation in the real world. Videos and source code are available at: https://sites.google.com/view/da-qd.

Emergence of massive dynamic objects will diversify spatial structures when robots navigate in urban environments. Therefore, the online removal of dynamic objects is critical. In this paper, we introduce a novel online removal framework for highly dynamic urban environments. The framework consists of the scan-to-map front-end and the map-to-map back-end modules. Both the front- and back-ends deeply integrate the visibility-based approach and map-based approach. The experiments validate the framework in highly dynamic simulation scenarios and real-world dataset.

This paper presents a novel manipulation method utilizing dynamic deformation of a flexible body with a structural anisotropy. Employing a spiral flexible body, a dynamic underactuated manipulation using its various vibrational patterns is proposed. First, the orbit of the tip of flexible body for the vibrational input to its root is theoretically derived. Subsequently, for flexible bodies with and without the structural anisotropy, structural stiffness and vibrational orbit of the tip of body are analyzed. Through this analysis, the generation mechanism of the orbit change effect according to the input frequency is revealed. Finally, the proposed method is experimentally validated. After confirming the orbit change effect in a spiral flexible body, this effect is applied to an underactuated nonprehensile manipulation where three-Dof motion of an object is controlled by a single actuator.

The objective of foraging robot swarms is to search for and collect resources in an unknown arena as quickly as possible. To avoid the congestion near the central collection zone, we previously proposed an extension to the multiple-place foraging in which robot chains are deployed dynamically so that foraging robots can deliver to the robot chains instead of the central collection zone. However, a robot chain can only reach one location at a time, and congestion can occur at the end of the robot chain. This paper presents an extension to dynamic robot chains called dynamic robot chain networks, which extends robot chains with branches, each of which reaches different resource clusters. We formulate the problem of finding the smallest dynamic robot chain networks as the Euclidean Steiner tree problem and explain how Steiner trees can be utilized to optimize the efficiency of the foraging operations. We implemented our foraging robot swarms in a simulator called ARGoS. Our experiments showed that dynamic robot chain networks can avoid obstacles and collect more resources when compared with the original robot chain design.

Collaborative robots are gradually taking over the leading position in automating the production and manufacturing of the SMEs, where the human-robot collaboration is highly emphasized. Therefore, estimating the force and simulating the performance of robots are of great importance. As a newly introduced technology, digital twin, has gained more attentions for simulation, process evaluation, real-time monitoring, etc. However, the current state-of-the-art of digital twin for robots still remains on the kinematic level, and the integrated robot system dynamics is too complex to be incorporated into the digital twin. Therefore, this research starts with the perspective of harmonic drive based robot joint, and proposes a dynamic model of robot joint by analyzing the composition, transmission principle, and internal interactions. Then the experimental parameter identification is performed to obtain the inherent parameters, which can reflect the system performance characteristics. Finally, a preliminary digital twin of robot joint integrated with dynamic model is established with Gazebo and MATLAB. The proposed approach could be used to simulate the dynamic behavior of robot joint in real time and make contributions to the state of the art for digital twin.

Recent works in Reinforcement Learning (RL) combine model-free (Mf)-RL algorithms with model-based (Mb)-RL approaches to get the best from both: asymptotic performance of Mf-RL and high sample-efficiency of Mb-RL. Inspired by these works, we propose a hierarchical framework that integrates online learning for the Mb-trajectory optimization with off-policy methods for the Mf-RL. In particular, two loops are proposed, where the Dynamic Mirror Descent based Model Predictive Control (DMD-MPC) is used as the inner loop Mb-RL to obtain an optimal sequence of actions. These actions are in turn used to significantly accelerate the outer loop Mf-RL. We show that our formulation is generic for a broad class of MPC based policies and objectives, and includes some of the well-known Mb-Mf approaches. We finally introduce a new algorithm: Mirror-Descent Model Predictive RL (M-DeMoRL), which uses Cross-Entropy Method (CEM) with elite fractions for the inner loop. Our experiments show faster convergence of the proposed hierarchical approach on benchmark MuJoCo tasks. We also demonstrate hardware training for trajectory tracking in a 2R leg, and hardware transfer for robust walking in a quadruped. We show that the inner-loop Mb-RL significantly decreases the number of training iterations required in the hardware setting, thereby validating the proposed approach.

Even though cobots have high potential in bringing several benefits in the manufacturing and logistic processes, their rapid (re-)deployment in changing environments is still limited. To enable fast adaptation to new product demands and to boost the fitness of the human workers to the allocated tasks, we propose a novel method that optimizes assembly strategies and distributes the effort among the workers in human-robot cooperative tasks. The cooperation model exploits AND/OR Graphs that we adapted to solve also the role allocation problem. The allocation algorithm considers quantitative measurements that are computed online to describe human operators' ergonomic status and task properties. We conducted preliminary experiments to demonstrate that the proposed approach succeeds in controlling the task allocation process to ensure safe and ergonomic conditions for the human worker.

This paper deals with the control of a redundant robotic system for middle ear surgery (i.e., cholesteatoma tissues removal). The targeted robotic system is a macro-micro-scale robot composed of a redundant seven degrees of freedom (DoFs) on which is attached a two DoFs robotized flexible fiberscope. Two different control architectures are proposed to achieve a defined surgical procedure to remove the pathological tissue inside the middle ear cavity. The first proposed control mode is based on the position-based tele-operation of the entire system using a joystick (Phantom Omni) as a master arm. The second one combines comanipulation of the seven DoFs robotic arm using an embedded force/torque sensor and an end-frame tele-operation of the remaining two DoFs fiberscope using a lab-made in-hand joystick. Experimental validation is performed to evaluate and compare the performance of both developed control schemes. The obtained results using the lab-made platform and the proposed controllers are discussed.

Hand pose estimation constitutes prime attainment for human-machine interaction-based applications. Real-time operation is vital in such tasks. Thus, a reliable estimator should exhibit low computational complexity and high precision at the same time. Previous works have explored the regression techniques, including the coordinate regression and heatmap regression methods. Primarily incorporating ideas from them, in this paper, we propose a novel, fast and accurate method for hand pose estimation, which adopts a lightweight network architecture and a post-processing scheme. Hence, our architecture uses a Dual Regression strategy, consisting of two regression branches, namely the coordinate and the heatmap ones, and we refer to the proposed method as DRHand. By carefully selecting the branches' characteristics, the proposed structure has been designed to exploit the benefits of the two methods mentioned above while impoverishing their weaknesses to some extent. The two branches are supervised separately during training, and a post-processing module estimates their outputs to boost reliability. This way, our novel pipeline is considerably faster, reaching 44.39 frames-per-second on an NVIDIA Jetson TX2 graphics processing unit, offering a beyond real-time performance for any custom robotics application. Lastly, extensive experiments conducted on two publicly-available datasets demonstrate that the proposed framework outperforms previous state-of-the-art techniques and can generalize on various hand pose scenarios.

Implementing and conducting reproducible experiments on multi-robot hardware platforms are challenging tasks due to variations in hardware, software, and most importantly the intensive implementation effort. In this paper, we aim to present the Driving Swarm software framework which is developed to facilitate the implementation, deployment, supervision, and analysis of multi-robot experiments. We use this framework with the TurtleBot3 hardware platform and measure its performance for two example scenarios: trajectory tracking and flocking behavior, with 5 and 6 robots. The goal of our experiments is to validate the simulation comparing to hardware implementations and to provide a baseline data for further experiments. While the simulated and real robots show similar behavior, we could observe that the simulated behavior is more robust than the real behavior in both scenarios. This effect is observed in the lower tracking error and better obstacle avoidance in both experiments. While the simulation proved to be a valuable tool during the development of the behaviors, the results confirm the importance of conducting experiments on a real-world test bed.

One of the challenges in vision-based driving trajectory generation is dealing with out-of-distribution scenarios. In this paper, we propose a domain generalization method for vision-based driving trajectory generation for autonomous vehicles in urban environments, which can be seen as a solution to extend the Invariant Risk Minimization (IRM) method in complex problems. We leverage an adversarial learning approach to train a trajectory generator as the decoder. Based on the pre-trained decoder, we infer the latent variables corresponding to the trajectories, and pre-train the encoder by regressing the inferred latent variable. Finally, we fix the decoder but fine-tune the encoder with the final trajectory loss. We compare our proposed method with the state-of-the-art trajectory generation method and some recent domain generalization methods on both datasets and simulation, demonstrating that our method has better generalization ability. Our project is available at https://sites.google.com/view/dg-traj-gen.

Robust imitation learning using disturbance injections overcomes issues of limited variation in demonstrations. However, these methods assume demonstrations are optimal, and that policy stabilization can be learned via simple augmentations. In real-world scenarios, demonstrations are often of diverse-quality, and disturbance injection instead learns sub-optimal policies that fail to replicate desired behavior. To address this issue, this paper proposes a novel imitation learning framework that combines both policy robustification and optimal demonstration learning. Specifically, this combinatorial approach forces policy learning and disturbance injection optimization to focus on mainly learning from high task achievement demonstrations, while utilizing low achievement ones to decrease the number of samples needed. The effectiveness of the proposed method is verified through experiments using an excavation task in both simulations and a real robot, resulting in high-achieving policies that are more stable and robust to diverse-quality demonstrations. In addition, this method utilizes all of the weighted sub-optimal demonstrations without eliminating them, resulting in practical data efficiency benefits.

Autonomous navigation of mobile robots is a well-studied problem in robotics. However, the navigation task becomes challenging when multi-robot systems have to cooperatively navigate dynamic environments with deadlock-prone layouts. We present a Distributed Timed Elastic Band (DTEB) Planner that combines Prioritized Planning with the online TEB trajectory Planner, in order to extend the capabilities of the latter to multi-robot systems. The proposed planner is able to reactively avoid imminent collisions as well as predictively resolve potential deadlocks among a team of robots, while navigating in a complex environment. The results of our simulation demonstrate the reliable performance and the versatility of the planner in different environment settings. The code and tests for our approach are available online.

Potential field approaches have been often used to describe and model interactions within a swarm of robots performing collective motion, also called flocking. Despite the high number of proposed approaches, most have only been tested in simulation and among the minority tested on real robots, even fewer abandoned the laboratory boundaries in favor of real-world scenarios. In this work, we propose a decentralized flocking approach that builds over the classical potential field models and that is proved to work well both in simulated and real-world environments. Each robot in the swarm relies on limited information and can only perceive its local neighbors through limited communication of noisy position information. No information on individual drone orientations, velocities, or accelerations is exchanged or needed. The novel experimental achievement of this paper is the realization of collective motion in three dimensions with the above sensing limitations. The swarm dynamically adapts to the environment by keeping a preferred distance from the ground and by changing formation. To show the general applicability of the proposed control algorithm, we study how it performs with the use of different potential functions proposed in the literature and by comparing them via extensive evaluation of the results in a realistic simulated environment. Lastly, we compare the performances of the proposed approach and of the different potentials on a real-drone swarm of up to fourteen robots flying both in two and three dimensional formations and in a challenging outdoor environment.

For aerial swarms, navigation in a prescribed formation is widely practiced in various scenarios. However, the associated planning strategies typically lack the capability of avoiding obstacles in cluttered environments. To address this deficiency, we present an optimization-based method that ensures collision-free trajectory generation for formation flight. In this paper, a novel differentiable metric is proposed to quantify the overall similarity distance between formations. We then formulate this metric into an optimization framework, which achieves spatial-temporal planning using polynomial trajectories. Minimization over collision penalty is also incorporated into the framework, so that formation preservation and obstacle avoidance can be handled simultaneously. To validate the efficiency of our method, we conduct benchmark comparisons with other cutting-edge works. Integrated with an autonomous distributed aerial swarm system, the proposed method demonstrates its efficiency and robustness in real-world experiments with obstacle-rich surroundings11https://www.youtube.com/watch?v=lFumtOrJci4. We will release the source code for the reference of the community22https://github.com/ZJU-FAST-Lab/Swarm-Formation.