Announced the expansion of its leerbot platform on Wednesday with a large dataset for the purpose of automotive automation. Online Artificial Intelligence (AI) and Machine Learning (ML) Repository said that the dataset was made in collaboration with AI Startup Yak. Dubbed Learning to Drive (L2D), dataset was collected from a suit of sensors installed on 60 electric vehicles (EVS) over a period of three years. The objective of the open-source dataset is to enable developers and robotics community to create spatial intelligence solutions for the automobile industry.
Hugging Face L2D adds dataset to Lerobot
One in blog postThe company expanded the new AI dataset, called it “the world’s largest multimodal dataset for the purpose of creating an open-vault spatial intelligence for the automotive domain.” The entire dataset is more than 1PB (a petabite) in size, and was collected using a censor suit installed on 60 EVs operated by driving schools in 30 German cities for three years. The same sensor was used to ensure stability in the data collected.
The Lerobot platform was launched as a collection of Open-SOS AI model, dataset and tools that can help developers to create AI-operated robotics systems last year.
Learning to run a dataset
Photo Credit: Hug face
The policies in the dataset are divided into two groups of expert policies and student policies. Earlier, data from driving trainers involves data while later comes from learner drivers. Hugging Face said that specialist policy has zero driving mistakes and is considered optimal, while the student policy has a sub-presence known in the student policy. Both groups include natural language instructions for driving tasks.
Each group has all driving landscapes that are required to complete the driving license in the European Union (EU). Some of these driving works include overtaking, roundabout handling and track driving.
Expanding the sensor suit used to capture the L2D data, Hugging Face said that each of the 60 Kia Nero EV model was equipped with six RGB cameras, which to catch the vehicle around the vehicle around the vehicle around the vehicle around the vehicle around the vehicle around the vehicle for a vehicle (IMU) to catch the dynamics of the vehicle. All data was captured with a timestamp.
In particular, the dataset aims to help developers and robotics scientists, who manufacture end-to-end self-driving AI models, which can eventually be used for the manufacture of completely autonomous vehicle systems.
Hugging Face highlighted that the L2D dataset would be released in a phased manner, where each gradual release would be a superset of previous release to ensure ease of access. The platform is also inviting the community to present models for a dataset’s closed loop test with a security driver. It will start in summer in 2025.
