A5087318700- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- Robotic Path Planning Algorithms
- Indoor and Outdoor Localization Technologies
- Modular Robots and Swarm Intelligence
- Advanced Vision and Imaging
- Optimization and Search Problems
- UAV Applications and Optimization
- Data Management and Algorithms
- Reinforcement Learning in Robotics
- Robotics and Automated Systems
- Evolutionary Algorithms and Applications
- Mobile Agent-Based Network Management
- Advanced Neural Network Applications
- Design Education and Practice
- Geographic Information Systems Studies
- 3D Surveying and Cultural Heritage
- Multimodal Machine Learning Applications
- Chaos, Complexity, and Education
- Distributed systems and fault tolerance
- Algorithms and Data Compression
- Distributed Control Multi-Agent Systems
- Architecture and Computational Design
- Autonomous Vehicle Technology and Safety
ETH Zurich
2015-2022
University of Zurich
2016-2022
Corvallis Environmental Center
2020
Robotics Research (United States)
2016
École Polytechnique Fédérale de Lausanne
2014
Exploring and mapping previously unknown environments while avoiding collisions with obstacles is a fundamental task for autonomous robots. In scenarios where this needs to be done rapidly, multi-rotors are good choice the task, as they can cover ground at potentially very high velocities. Flying velocities, however, implies ability rapidly plan trajectories react new information quickly. paper, we propose an extension classical frontier-based exploration that facilitates speeds. The...
Decentralized visual simultaneous localization and mapping (SLAM) is a powerful tool for multi-robot applications in environments where absolute positioning not available. Being visual, it relies on cheap, lightweight versatile cameras, and, being decentralized, does rely communication to central entity. In this work, we integrate state-of-the-art decentralized SLAM components into new, complete system. To allow data association optimization, existing systems exchange the full map among all...
Despite impressive results in visual-inertial state estimation recent years, high speed trajectories with six degree of freedom motion remain challenging for existing algorithms. Aggressive feature large accelerations and rapid rotational motions, when they pass close to objects the environment, this induces apparent motions vision sensors, all which increase difficulty estimation. Existing benchmark datasets do not address these types trajectories, instead focusing on slow or constrained...
Transport of objects is a major application in robotics nowadays. While ground robots can carry heavy payloads for long distances, they are limited rugged terrains. Aerial deliver arbitrary terrains; however tend to be payload. It has been previously shown that, payloads, it beneficial them using multiple flying robots. In this paper, we propose novel collaborative transport scheme, which two quadrotors cable-suspended payload at accelerations that exceed the capabilities previous...
This paper presents a novel system for autonomous, vision-based drone racing combining learned data abstraction, nonlinear filtering, and time-optimal trajectory planning. The has successfully been deployed at the first autonomous world championship:
Robust, scalable place recognition is a core competency for many robotic applications. However, when revisiting places over and over, state-of-the-art approaches exhibit reduced performance in terms of computation memory complexity accuracy. For successful deployment robots long time scales, we must develop algorithms that get better with repeated visits to the same environment, while still working within fixed computational budget. This paper presents evaluates an algorithm alternates...
This paper presents a novel system for autonomous, vision-based drone racing combining learned data abstraction, nonlinear filtering, and time-optimal trajectory planning.The has successfully been deployed at the first autonomous world championship: 2019 AlphaPilot Challenge.Contrary to traditional systems, which only detect next gate, our approach makes use of any visible gate takes advantage multiple, simultaneous detections compensate drift in state estimate build global map gates.The...
Robotic practitioners generally approach the vision-based SLAM problem through discrete-time formulations. This has advantage of a consolidated theory and very good understanding success failure cases. However, needs tailored algorithms simplifying assumptions when high-rate and/or asynchronous measurements, coming from different sensors, are present in estimation process. Conversely, continuous-time SLAM, often overlooked by practitioners, does not suffer these limitations. Indeed, it...
Science instructors from a wide range of disciplines agree that hands-on laboratory components courses are pedagogically necessary (Freedman, 1997). However, certain shortcomings current exercises have been pointed out by several authors (Mataric, 2004; Hofstein and Lunetta, 2004). The overarching theme these analyses is tend to be formulaic, closed-ended, at times outdated. To address issues, we envision novel platform not only didactic tool but also an experimental testbed for users play...
Place recognition is a core component in simultaneous localization and mapping (SLAM), limiting positional drift over space time to unlock precise robot navigation. Determining which previously visited places belong together continues be highly active area of research as robotic applications demand increasingly higher accuracies. A large number place algorithms have been proposed, capable consuming variety sensor data including laser, sonar depth readings. The best performing solutions,...
State-of-the-art systems that place recognition in a group of n robots either rely on centralized solution, where each robot's map is sent to central server, or decentralized the all other robots, within communication range. Both approaches have their drawbacks: entity, which handles computational load and cannot be deployed large, remote areas, whereas exchange times more data preclude matches between visit same at different while never being close enough communicate directly. We propose...
Persistent merging of maps created by different sensor modalities is an insufficiently addressed problem. Current approaches either rely on appearance-based features which may suffer from lighting and viewpoint changes or require pre-registration between all used. This work presents a framework using structural descriptors for matching LIDAR point-cloud sparse vision keypoint maps. The algorithm works independently the sensors' varying does not sensors Furthermore, we employ approach in...
In this paper, we propose to augment image-based place recognition with structural cues. Specifically, these cues are obtained using structure-from-motion, such that no additional sensors needed for recognition. This is achieved by augmenting the 2D convolutional neural network (CNN) typically used a 3D CNN takes as input voxel grid derived from structure-from-motion point cloud. We evaluate different methods fusing and features obtain best performance global average pooling simple...
Large scale, long-term, distributed mapping is a core challenge to modern field robotics. Using the sensory output of multiple robots and fusing it in an efficient way enables creation globally accurate consistent metric maps. To combine data from agents into global map, most existing approaches use central entity that collects manages information all agents. Often, raw sensor one robot needs be made available processing algorithms on other due lack computational resources robot....
Large scale, long-term, distributed mapping is a core challenge to modern field robotics. Using the sensory output of multiple robots and fusing it in an efficient way enables creation globally accurate consistent metric maps. To combine data from agents into global map, most existing approaches use central entity that collects manages information all agents. Often, raw sensor one robot needs be made available processing algorithms on other due lack computational resources robot....
In this paper, we discuss the adaptation of our decentralized place recognition method described in [1] to full image descriptors. As had shown, key making a scalable visual lies exploting deterministic assignment distributed key-value map. Through this, it is possible reduce bandwidth by up factor n, robot count, casting lookup problem. [1], exploited for bag-of-words [3], [4]. Our bag-of-words, however, results complex system, which has inherently worse recall than its centralized...
Science instructors from a wide range of disciplines agree that hands-on laboratory components courses are pedagogically necessary (Freedman, 1997). However, certain shortcomings current exercises have been pointed out by several authors (Mataric, 2004; Hofstein and Lunetta, 2004). The overarching theme these analyses is tend to be formulaic, closed-ended, at times outdated. To address issues, we envision novel platform not only didactic tool but also an experimental testbed for users play...
A wide range of computer vision algorithms rely on identifying sparse interest points in images and establishing correspondences between them. However, only a subset the initially identified results true (inliers). In this paper, we seek detector that finds minimum number are likely to result an application-dependent "sufficient" inliers k. To quantify goal, introduce "k-succinctness" metric. Extracting is attractive for many applications, because it can reduce computational load, memory,...
For robotics and augmented reality systems operating in large dynamic environments, place recognition tracking using vision represent very challenging tasks. Additionally, when these need to reliably operate for long time periods, such as months or years, further challenges are introduced by severe environmental changes, that can significantly alter the visual appearance of a scene. Thus, unlock term, large scale place recognition, it is necessary develop new methodologies...
The extraction and matching of interest points is a prerequisite for many geometric computer vision problems. Traditionally, has been achieved by assigning descriptors to that have similar descriptors. In this paper, we propose method which are instead already implicitly matched at detection time. With this, do not need be calculated, stored, communicated, or any more. This convolutional neural network with multiple output channels can thought as collection variety detectors, each...