ICRA 2021 论文列表

Collision detection and recovery for aerial robots remain a challenge because of the limited space for sensors and local stability of the flight controller. We introduce a novel collision-resilient quadrotor that features a compliant arm design to enable free flight while allowing for one passive degree of freedom to absorb shocks. We further propose a novel collision detection and characterization method based on Hall sensors, as well as a new recovery control method to generate and track a smooth trajectory after a collision occurs. Experimental results demonstrate that the robot can detect and recover from high-speed collisions with various obstacles such as walls and poles. Moreover, it can survive collisions that are hard to detect with existing methods based on IMU data and contact models, for example, when colliding with unstructured surfaces, or being hit by a moving obstacle while hovering.

Knowledge of interaction forces during teleoperated robot-assisted surgery could be used to enable force feedback to users and evaluate tissue handling skill. However, direct force sensing at the end-effector is challenging because it requires biocompatible, sterilizable, and cost-effective sensors. Vision-based neural networks are a promising approach for providing useful force estimates, though questions remain about generalization to new scenarios and real-time inference. We present a force estimation neural network that uses RGB images and robot state as inputs. Using a self-collected dataset, we compared the network to variants that included only a single input type, and evaluated how they generalized to new viewpoints, workspace positions, materials, and tools. We found that the vision-only network was sensitive to shifts in viewpoints, while networks with state inputs were sensitive to vertical shifts in workspace. The network with both state and vision inputs had the highest accuracy for an unseen tool, while the state-only network was most accurate for an unseen material. Through feature removal studies, we found that using only force features produced better accuracy than using only kinematic features as input. The network with both state and vision inputs outperformed a physics-based model in accuracy for seen material. It showed comparable accuracy but faster computation times than a recurrent neural network, making it better suited for real-time applications.

We present fully autonomous source seeking onboard a highly constrained nano quadcopter, by contributing application-specific system and observation feature design to enable inference of a deep-RL policy onboard a nano quadcopter. Our deep-RL algorithm finds a high-performance solution to a challenging problem, even in presence of high noise levels and generalizes across real and simulation environments with different obstacle configurations. We verify our approach with simulation and in-field testing on a Bitcraze CrazyFlie using only the cheap and ubiquitous Cortex-M4 microcontroller unit. The results show that by end-to-end application-specific system design, our contribution consumes almost three times less additional power, as compared to a competitive learning-based navigation approach onboard a nano quadcopter. Thanks to our observation space, which we carefully design within the resource constraints, our solution achieves a 94% success rate in cluttered and randomized test environments, as compared to the previously achieved 80%. We also compare our strategy to a simple finite state machine (FSM), geared towards efficient exploration, and demonstrate that our policy is more robust and resilient at obstacle avoidance as well as up to 70% more efficient in source seeking. To this end, we contribute a cheap and lightweight end- to-end tiny robot learning (tinyRL) solution, running onboard a nano quadcopter, that proves to be robust and efficient in a challenging task.

Obtaining highly dynamic motions in robots is a difficult task. In recent years, sophistication in mechanical design, improved algorithms, and high computational power allows new robots to perform natural gaits and dynamic motions such as backflips. Offline optimization is often necessary to obtain good performance in those difficult motions. However, when an athlete does a backflip, he will adapt “online” to any change, and that is shown in the robustness of the movements. One of the biggest challenges in robotics is to perform those movements using online optimization with the dynamics of the robot. Here, we present an approach to deal with complicated tasks using online optimization. We obtain 90° rotational jumps and jumps over sloped terrain in the Mini-Cheetah hardware, and online-optimized backflips, sideflips, and frontflips in a real-time physical simulator with full-body dynamics.

In this paper, we investigate the problem of generating time-optimal trajectories automatically for spraying pesticides to infected regions with varying degree of infections in an agricultural field using a collection of quad-rotors that have limited pesticide carrying capacity but are capable of refilling from the pesticide tanks stationed across the field. We consider the time of traversal between points, time for changing the direction of traversal, and time for spraying and refilling, explicitly in the problem formulation, and generate trajectories for the multiple quad-rotors to execute in parallel such that the total energy (time) for completion of the task is minimized. We present two methods namely, multiple traveling salesman based time-optimal trajectory generation, and clustering based de-compositional time-optimal trajectory generation, respectively. In the first method, our approach consists of reducing the time-optimal trajectory generation problem for the multiple quad-rotors to a version of the multiple traveling salesman problem on a weighted graph, and then encoding the problem into a mixed integer linear programming problem. In the second method, our approach consists of first decomposing the infected regions into k clusters corresponding to the k robots, placing each quad-rotor at the center of the corresponding clusters, and finally, independently solving time-optimal trajectory generation problem for each quad-rotor using the first method. We have implemented both the approaches and performed experimental comparison. We observe that the first method provides an optimal strategy, while the second method provides a sub-optimal strategy. However, the second method is computationally more efficient, and the loss of optimality is minimal, hence, it provides a better trade-off between the computational complexity and the optimality constraints.

The time-domain passivity approach has been proposed in the literature in a variety of formats to guarantee the stability of teleoperation leader-follower systems. The conventional use of the proposed technique utilizes the control effort at the follower side as the force feedback to be sent back to the user at the leader’s side. However, this has resulted in transparency problems, especially when the follower dynamics are not negligible. On the other hand, four-channel and three-channel Lawrence architectures have been investigated widely in the literature to maximize the transparency of the system when, in most advanced cases, stability is guaranteed using wave-variables. However, wave-variables are historically known for their transparency deterioration problems. In this paper, we propose a two-layer approach taking advantage of the fusion of (a) a more optimal derivation of Lawrence telerobotic architecture (utilizing only two channels), and (b) a two-port time-domain passivity stabilizer while comparing the performance with a one-port passivity stabilizer. The two-channel derivation of the Lawrence architecture allows for implementing a two-port time domain passivity approach, which is investigated in this paper. The performance of this is compared systematically through a multi-objective approach by analyzing dissipated energy besides force and velocity errors for a wide range of time delays and frequencies of excitation. The paper gives a comprehensive view of the efficacy of two-port versus one-port time-domain passivity control when combined with the two-channel derivation of Lawrence architecture.

Planning the simultaneous movement of multiple agents represents a challenging coordination problem, and ideally safety and efficiency are jointly addressed. This paper introduces a planning algorithm for fast and energy-efficient trajectories with reduced collision potential from a start to an end constellation. This new approach combines trajectory approximation based on model predictive control, collision avoidance with potential fields, and flight energy optimization with minimum snap trajectories. Our approach results in unprecedented transition times and success rates with less energy consumption, as shown in simulation and real experiments with 16 drones.

In unstructured and unknown environments, heterogeneous robots must be able to perceive the environment, coordinate with each other and complete tasks collaboratively with onboard sensors. In this paper, a tightly-coupled perception and navigation framework is proposed for heterogeneous land-air robots, which forms a closed loop of perception-navigation for heterogeneous robots. The key novelty of this work is the proposing of a unified framework to formulate the cooperative mapping and navigation problem, as well as the derivation of high-level coordination strategy and low-level goal-oriented navigation within a fully integrated approach. To provide a comprehensive understanding of the environment, a flexible probabilistic map fusion algorithm is applied to merge local maps generated by hybrid robots. The proposed UAV-UGV hybrid system is validated in challenging experiments, proving its robustness and effectiveness in practical tasks.

Robust and accurate pose estimation in long-term localization is crucial to autonomous driving. In this paper, we dealt with absolute localization with a LiDAR feature map and multi-sensor measurements. We proposed a tightly-coupled fusion method with fixed-lag smoothing. A sliding window of recently maintained states is estimated by minimizing a joint cost function. This cost function includes residuals of global LiDAR registration and relative kinematic constraints from an IMU and wheel encoders. In addition, we enhance the robustness of our method by improving LiDAR registration. To achieve this goal, LiDAR feature maps with a hybrid of geometric and normal distribution features are constructed and exploited. The effectiveness of the proposed method is verified in several challenging test sequences over 200km. The experimental results demonstrate that the proposed method achieves accurate localization and high robustness in challenging scenarios even when the LiDAR observation is degraded.

In this paper we propose a framework for inte-grating map-based relocalization into online direct visual odometry. To achieve map-based relocalization for direct methods, we integrate image features into Direct Sparse Odometry (DSO) and rely on feature matching to associate online visual odometry (VO) with a previously built map. The integration of the relocalization poses is threefold. Firstly, they are incorporated as pose priors in the direct image alignment of the front-end tracking. Secondly, they are tightly integrated into the back-end bundle adjustment. Thirdly, an online fusion module is further proposed to combine relative VO poses and global relocalization poses in a pose graph to estimate keyframe-wise smooth and globally accurate poses. We evaluate our method on two multi-weather datasets showing the benefits of integrating different handcrafted and learned features and demonstrating promising improvements on camera tracking accuracy.

In the Wave-driven unmanned surface vehicles (WUSVs), oscillating-foils are the most straightforward and widely used wave energy conversion mechanism. In this paper, a kind of novel asymmetric foil is proposed, which improves the wave energy-converting efficiency to provide a more significant thrust in every wave cycle. We break down the movement of the foils in the wave and build the corresponding kinetic model to analyze their working effectiveness numerically. Through computational fluid dynamic (CFD) simulations, we determine the optimal values of critical parameters of the foils, which are suitable for a wide range of wave conditions. The thrust enhancement of the asymmetric foil is verified in both CFD simulations and hydrodynamic experiments, and the result shows a similar enhancement trend. Comparing with the traditional symmetric foil, our asymmetric foil can provide at least 13.75% more thrust to the WUSVs.

Despite the advances in autonomous navigation and motion planning, there are still several challenges to overcome, especially for confined or underground spaces. Confined scenarios present challenges such as lack of global or accurate external localization, uneven and slippery terrains, and multilevel stages. Exploring and mapping unknown unstructured environments is a fundamental step into the safe and efficient accomplishment of real-world tasks such as search and rescue missions or the autonomous inspection of dangerous areas. This paper proposes a novel three-dimensional autonomous exploration method for ground robots that considers the terrain traversability combined with the frontier expected information gain as a metric for the next best frontier selection in GPS-denied, confined spaces. Safe paths for navigation and frontier extraction are calculated iteratively from multiple 3D map representations such as octrees and meshes. Results in realistic simulated underground scenarios from the DARPA subterranean challenge demonstrate the technique's feasibility, achieving a more reliable and faster exploration rate over competing approaches.

ICSI (Intracytoplasmic sperm injection) is one of the most effective treatments for severe male infertility. During the implementation of the ICSI, it is necessary to perform the three-dimensional positioning of the tip of the glass injection micropipette. At present, the process is mainly controlled by skilled operators. Such manual operation is time-consuming and likely to cause micropipette damage or even serious damage to the oocyte membrane due to the inaccurate positioning. In this paper, an automatic system was developed for micropipette positioning. The microscopic positioning is carried out by autofocusing and planar positioning. A search strategy based on Dynamic Curve Fitting (DCF) was proposed to avoid local extremum issues due to the noises during the autofocusing. The proposed search strategy is adaptive to different focus algorisms, making the design of autofocusing becomes simpler. Besides, we proposed a planar positioning algorithm that can quickly and accurately obtain the tip position of the micropipette at the focusing plane. Finally, the visual servo control is employed to move the micropipette to the center of the vision. The experimental results demonstrated that the DCF method can accurately (mean error: 3 μm) find the focusing plane when applying random Gaussian noise with a mean value of 0 and a variance of 1 to the focus measure. By contrast, both the hill-climbing search method and the Fibonacci search method cannot work properly. The tip positioning algorithm provides a real-time in-plane tip positioning at a frame rate of 40 Hz with an average accuracy of 11 μm.

Across a wide range of applications, from autonomous vehicles to medical imaging, multi-spectral images provide an opportunity to extract additional information not present in color images. One of the most important steps in making this information readily available is the accurate estimation of dense correspondences between different spectra.Due to the nature of cross-spectral images, most correspondence solving techniques for the visual domain are simply not applicable. Furthermore, most cross-spectral techniques utilize spectra-specific characteristics to perform the alignment. In this work, we aim to address the dense correspondence estimation problem in a way that generalizes to more than one spectrum. We do this by introducing a novel cycle-consistency metric that allows us to self-supervise. This, combined with our spectra-agnostic loss functions, allows us to train the same network across multiple spectra.We demonstrate our approach on the challenging task of dense RGB-FIR correspondence estimation. We also show the performance of our unmodified network on the cases of RGB-NIR and RGB-RGB, where we achieve higher accuracy than similar self-supervised approaches. Our work shows that cross-spectral correspondence estimation can be solved in a common framework that learns to generalize alignment across spectra.

Simulating realistic radar data has the potential to significantly accelerate the development of data-driven approaches to radar processing. However, it is fraught with difficulty due to the notoriously complex image formation process. Here we propose to learn a radar sensor model capable of synthesising faithful radar observations based on simulated elevation maps. In particular, we adopt an adversarial approach to learning a forward sensor model from unaligned radar examples. In addition, modelling the backward model encourages the output to remain aligned to the world state through a cyclical consistency criterion. The backward model is further constrained to predict elevation maps from real radar data that are grounded by partial measurements obtained from corresponding lidar scans. Both models are trained in a joint optimisation. We demonstrate the efficacy of our approach by evaluating a down-stream segmentation model trained purely on simulated data in a real-world deployment. This achieves performance within four percentage points of the same model trained entirely on real data.

Industrial manipulators do not collapse under their own weight when powered off due to the friction in their joints. Although these mechanism are effective for stiff position control of pick-and-place, they are inappropriate for legged robots that must rapidly regulate compliant interactions with the environment. However, no metric exists to quantify the robot’s performance degradation due to mechanical losses in the actuators and transmissions. This paper provides a fundamental formulation that uses the mechanical efficiency of transmissions to quantify the effect of power losses in the mechanical transmissions on the dynamics of a whole robotic system. We quantitatively demonstrate the intuitive fact that the apparent inertia of the robots increase in the presence of joint friction. We also show that robots that employ high gear ratio and low efficiency transmissions can statically sustain more substantial external loads. We expect that the framework presented here will provide the fundamental tools for designing the next generation of legged robots that can effectively interact with the world.

Model Predictive Control (MPC) is a classic tool for optimal control of complex, real-world systems. Although it has been successfully applied to a wide range of challenging tasks in robotics, it is fundamentally limited by the prediction horizon, which, if too short, will result in myopic decisions. Recently, several papers have suggested using a learned value function as the terminal cost for MPC. If the value function is accurate, it effectively allows MPC to reason over an infinite horizon. Unfortunately, Reinforcement Learning (RL) solutions to value function approximation can be difficult to realize for robotics tasks. In this paper, we suggest a more efficient method for value function approximation that applies to goal-directed problems, like reaching and navigation. In these problems, MPC is often formulated to track a path or trajectory returned by a planner. However, this strategy is brittle in that unexpected perturbations to the robot will require replanning, which can be costly at runtime. Instead, we show how the intermediate data structures used by modern planners can be interpreted as an approximate value function. We show that that this value function can be used by MPC directly, resulting in more efficient and resilient behavior at runtime.

In this paper, we present an experiment, designed to investigate and evaluate the scalability and the robustness aspects of mobile manipulation. The experiment involves performing variations of mobile pick and place actions and opening/closing environment containers in a human household. The robot is expected to act completely autonomously for extended periods of time. We discuss the scientific challenges raised by the experiment as well as present our robotic system that can address these challenges and successfully perform all the tasks of the experiment. We present empirical results and the lessons learned as well as discuss where we hit limitations.

This paper describes Resh, a new, statically typed, interpreted programming language and associated runtime for orchestrating multirobot systems. The main features of Resh are: (1) It offloads much of the tedious work of programming such systems away from the programmer and into the language runtime; (2) It is based on a small set of temporal and locational operators; and (3) It is not restricted to specific robot types or tasks. The Resh runtime consists of three engines that collaborate to run a Resh program using the available robots in their current environment. This paper describes both Resh and its runtime and gives examples of its use.