Home PC News DuckieNet lets developers test autonomous vehicle systems using toy cars

DuckieNet lets developers test autonomous vehicle systems using toy cars

Robotics evaluation has a reproducibility downside, owing partly to robots’ myriad interacting parts. These parts are sometimes sophisticated, solely partially observable, and educated with AI strategies the place effectivity varies considerably all through environments. In an effort to cope with among the many challenges explicit to the autonomous driving space, researchers at ETH Zurich, the Toyota Technical Institute, Mila in Montreal, and NuTonomy developed what they title the Decentralized Urban Collaborative Benchmarking Network (DuckieNet), a setup constructed using the open provide Duckietown platform. DuckieNet provides a framework for rising, testing, and deploying every notion and navigation algorithms, and the researchers declare it’s extraordinarily scalable nonetheless low cost to assemble.

The Duckietown project, which was conceived by a 2016 graduate class at MIT, consists of low price wheeled robots often known as Duckiebots constructed nearly totally from off-the-shelf parts. The solely onboard sensor is the forward-facing digicam; a Raspberry Pi handles computation and a pair of DC motors power the wheels. Duckietowns are the roads, which might be constructed from practice mats and tape, and the signage the robots use to navigate spherical. Traffic lights have the an identical {{hardware}} as a result of the Duckiebots (excluding the wheels) and are in a position to sensing, computing, and actuation by means of their LEDs.

DuckieNet builds on Duckietown by together with specialised parts to the platform, A challenges server retailers machine learning algorithms, benchmarks, and outcomes and computes leaderboards, dispatching jobs to be executed to a set of research machines. The evaluation machines, which could be native or cloud-based, run autonomous driving simulations. Physical labs with DuckieNet installations carry out real-world experiments — a localization neighborhood of “watchtowers” (low-cost constructions that use the an identical sensing and computation as a result of the Duckiebots) tracks tags affixed to the Duckiebots’ our our bodies.


DuckieNet is in some strategies akin to AWS DeepRacer, Amazon’s service that gives builders with a cloud-based simulator for rising autonomous driving fashions and deploying them to a model car. But DuckieNet prospects can define benchmarks like suggest place deviation (lateral displacement of a Duckiebot from the center of a lane) and suggest orientation deviation (suggest orientation with respect to the lane orientation) in Docker containers submitted to the challenges server. (Algorithms will even be submitted as Docker containers and observed.) Moreover, excepting duties like resetting experiments and recharging the Duckiebots, the platform is totally autonomous.

The researchers say one in all DuckieNet’s key features is web internet hosting evaluation competitions. In degree of reality, DuckieNet has been used since early 2019 inside the AI Driving Olympics, a bi-annual rivals that serves to benchmark the state-of-the-art in autonomous car driving. DuckieNet generates visualizations of effectivity metrics and leaderboards whereas providing entry to the underlying raw information, along with open provide baselines and documentation.

“Our contention is that there is a need for stronger efforts towards reproducible research for robotics, and that to achieve this we need to consider the evaluation in equal terms as the algorithms themselves,” the researchers wrote in a paper describing their work. “In this fashion, we can obtain reproducibility by design through the research and development processes. Achieving this on a large-scale will contribute to a more systemic evaluation of robotics research and, in turn, increase the progress of development.”

Most Popular

Recent Comments