#ICRA2022 Competitions – Robohub

Picture credit: Sensible Owl Multimedia

As one of many ICRA Science Communication Award Winner, I coated the digital facets of ICRA 2022. IEEE Worldwide Convention on Robotics and Automation (ICRA) 2022 is completely the most effective robotics convention. It usually covers an enormous vary of robotics together with however not restricted to notion, management, optimization, machine studying and application-robotics. In 2022, ICRA was held in Philadelphia, the place the U.S. declaration of independence was signed, for every week from Could, twenty third to Could, twenty seventh. This convention can be one of many first in-person conferences for roboticists after a few pandemic years. The convention had 7876 registered contributors, out of which 4703 contributors attended the convention in-person. You’ll be able to entry the convention technical papers and presentation right here. There have been additionally workshops, competitions, plenary talks, boards and networking occasions. For extra particulars in regards to the convention, please confer with the convention official web site right here.

On account of journey points, I couldn’t attend ICRA 2022 in-person. Regardless, I’ve tried my finest to share my expertise as a presenter and a digital attendee. Whereas I can solely seize a few keypoints alongside the trajectory in the course of the restricted time, I hope they’re true positives and generate a exact reconstruction of ICRA expertise, from a first-time ICRA presenter’s perspective.

Competitions

ICRA 2022 had 10 main competitions organized all through the convention week. On this article, let’s take a fast take a look at what challenges in robotics have been addressed by way of the organized competitions:

The BARN Problem was designed for a robotic to navigate from a predefined begin pose to a aim pose with minimal time whereas avoiding collisions. The robotic used 2D LiDAR for notion and a microcontroller with a most velocity of 2m/s. Through the competitors, the computation of the robotic was restricted to Intel i3 CPU with 16GB of DDR4 RAM. The competitors primarily used simulated BARN dataset (Perille et al., 2020), which has 300 pre-generated navigation environments, starting from straightforward open areas to troublesome extremely constrained ones, and an surroundings generator to generate novel BARN environments. The competitors allowed the taking part groups to make use of any navigation approaches, starting from classical sampling-based, optimization-based, end-to-end studying, to hybrid approaches.

Common Place Recognition Competitors was designed to improve visible and LiDAR state-of-the-art strategies for localization in large-scale environments with altering circumstances reminiscent of variations in viewpoints and environmental circumstances (e.g. illumination, season, time of day). The competitors had two challenges primarily based on Metropolis-scale UGV Localization Dataset (3D-3D Localization) and Visible Terrain Relative Navigation Dataset (2D-2D Localization) to guage efficiency in each long-term and large-scale.

RoboMaster College Sim2Real Problem was designed to optimize the system efficiency in real-world. Contributors developed algorithms in a simulated surroundings and the organizers deployed the submitted algorithms in real-world. The competitors centered on system efficiency together with notion, manipulation and navigation of the robotic.

RoboMaster College AI Problem centered on the applying of a number of facets of cellular robotics algorithm in an built-in context reminiscent of localization, movement planning, goal detection, autonomous decision-making and computerized management. The thought of the competitors was for the robots to shoot towards one another within the rune-filled battlefield and to launch projectiles towards different robots.

F1TENTH Autonomous Racing was desinged as an in-person competitors anticipating contributors to construct 1:10 scaled autonomous race automobile in keeping with a given specification and as a digital competitors to work on the simulation surroundings. The paricipating groups constructed the algorithms to finish the duty with no collisions and potential minimal laptime. This competitors centered on engineering facets of robotics together with dependable {hardware} system and sturdy algorithms.

Robotic Greedy and Manipulation Competitions was designed as three tracks, open cloud robotic desk group problem (OCRTOC), service observe and manufacturing observe. OCRTOC (Liu et al., 2021) observe was desiged to make use of a benchmark developed for robotic greedy and manipulation (Solar et al., 2021). Because the benchmark focuses on the object rearrangement downside, the competitors centered on offering a set of an identical actual robotic setups and faciliated distant experiments of standardized desk group eventualities of various difficulties. Service observe as an alternative centered on a single job of setting a proper dinner desk together with setting down dinner plates, a bowl, a glass and a cup, inserting silverware and napkins across the plates and eventually filling a glass and cup. Manufacturing observe competitors was designed to carry out each meeting and disassembly of a NIST Taske Board (NTB) that had threaded fasteners, pegs of assorted geometries, electrical connectors, wire connections and rounting, and a versatile belt with a tensioner.

DodgeDrone Problem: Imaginative and prescient-based Agile Drone Flight was designed to grasp the wrestle in autonomous navigation to realize the agility, versatility and robustness of people and animals, and to incentivize and facilitate analysis on this matter. The contributors developed notion and management algorithms to navigate a drone in each static and dynamic environments, and the organizers additionally supplied the contributors with an easy-to-use API and a reinforcement studying framework.

RoboJawn FLL Problem was designed just like conventional LEGO League occasion throughout which taking part groups competed with their robots in three CARGO CONNECT marches, and have been judged primarily based on innovation and robotic design.

SeasonDepth Prediction Problem centered on coping with long-term robustness of notion beneath varied environments for lifelong reliable autonomy within the utility of outside cellular robotics and autonomous driving. This competitors was the primary open-source problem specializing in depth prediction efficiency beneath totally different environmental circumstances and was primarily based on a monocular depth prediction dataset, SeasonDepth (Hu et al., 2021). There have been two tracks supervised studying observe and self-supevised studying observe with 7 slices of coaching set every beneath 12 totally different environmental circumstances.

Roboethics Competitors centered on designing robots to navigate ethically delicate conditions, like for instance, if a customer requests a robotic to fetch the house owner’s bank card, how ought to the robotic react or what iss the correct reply to an underaged teenager asking for an alcoholic drink. The Roboethics Competitors challenged groups at a hackathon occasion to design robots in a simulated surroundings that may navigate these difficult conditions in house. There was additionally one other observe of ethics problem, an answer by way of quick video presentation and challenge report, which have been then applied throughout hackathon.

References

1. Perille, D., Truong, A., Xiao, X. and Stone, P., 2020. Benchmarking Metric Floor Navigation. Worldwide Symposium on Security, Safety and Rescue Robotics (SSRR).
2. Solar, Y., Falco, J., Roa, M. A. and Calli, B., 2021. Analysis challenges and progress in robotic greedy and manipulation competitions. Robotics and Automation Letters, 7(2), 874-881.
3. Liu, Z., Liu, W., Qin, Y., Xiang, F., Gou, M., Xin, S., Roa, M. A. and Calli, B., Su, H., Solar Y. and Tan, P., 2021. Analysis challenges and progress in robotic greedy and manipulation competitions. Robotics and Automation Letters, 7(1), 486-493.
4. Hu, H., Yang, B., Qiao, Z., Zhao, D. and Wang, H., 2021. SeasonDepth: Cross-Season Monocular Depth Prediction Dataset and Benchmark beneath A number of Environments.

tags: c-Occasions