In order to promote the development of behavior prediction technology in the field of autonomous driving and accelerate the research of prediction model/algorithm evaluation, The University of California, Berkeley’s Mechanical Systems and Control Lab (MSC Lab), together with Graviti, an AI data service platform provider, and AWS, the world’s leading cloud service provider, officially launched the Interaction Dataset Prediction Challenge.
At present, the academia and industry agree that the behavioral Prediction (Prediction: such as trajectory, action, intention) is one of the most challenging problems in the field of automatic driving, and it is a major factor hindering the realization of fully automatic driving. In order to solve this problem, there are two indispensable conditions: first, the collection and accumulation of real scene motion data including many vehicle and pedestrian interactions; second, the correct and effective evaluation of various prediction algorithms can be carried out through these data. Unfortunately, there is no benchmark for fairly comparing the performance of different prediction models (or algorithms), especially when you consider that there is a programming loop (the model integrates both prediction and planning modules), and a fair benchmark is a big problem.
To this end, the MSC Lab at the University of California, Berkeley, and collaborators from the Karlsruhe Institute of Technology (KIT) and the Mines Paristech have established an international, admissible, and collaborative data set (Interaction). It can accurately reproduce a large number of interactive behaviors of road users (such as vehicles and pedestrians) in various driving scenarios in different countries.
To accelerate discussion and research in academia and industry around the evaluation of predictive models/algorithms, the Mechanical Systems Control Laboratory at the University of California, Berkeley, in collaboration with AI data service platform provider Graviti and AWS launched the Prediction Challenge based on the Interaction data set (Explt). The challenge aims to build effective and valuable predictive methods for the development of autonomous driving.
Challenge sponsor Graviti is committed to making machine learning simple, enabling all creative individuals and businesses to easily use AI and make it within reach. Graviti is willing to help more developers make driverless cars a reality faster through its one-stop machine learning platform.
The Graviti data set management system will be used to host the training, validation and testing data used in the competition, and will be made available to participants for download. At the same time, Graviti’s model evaluation framework is used to calculate the results achieved by competitors uploading results and generate leaderboards. AWS will provide cloud service resources and support for all services.
The Interaction Dataset Prediction Challenge is divided into five challengesGeneral Challenge, Data Efficiency Challenge, Generality Challenge, Closed Loop Challenge, Open Source Challenge. The competition is planned inTwo rounds are held in March and JuneAccording to the relatively fair index, the prediction model and algorithm are evaluated and tested in different aspects. Competitors can obtain inputs and outputs from the training and validation data sets, but only from the test set. After participants submit the results of the algorithm output on the test set, a series of evaluations will be conducted and scores will be posted on the activity leaderboard.The first round runs until the end of May, and the challenge data set and results will be announced at the Waymo 2020 CVPR Forum in June.
Participants are welcome to try different challenges.
Challenge Chinese Entry:http://challenge-zh.interaction-dataset.com
About: Mechanical Systems and Control Laboratory (MSC Laboratory), University of California, Berkeley
The Mechanical Systems and Control Laboratory (MSC Laboratory) at the University of California, Berkeley is headed by Professor Massayoshi Tomizuka. Professor Massayoshi Tomizuka joined the Department of Mechanical Engineering at UCB in 1974 and is currently Cheryl and John Neerhout, Jr. Distinguished professor. The MSC Lab has a history of more than 30 years of research in the field of automotive automation and control, and collaborated with PATH, California, to present Demo’97.
Over the past decade, the MSC lab has focused on intelligent/autonomous systems and their interactions with humans, in the areas of manufacturing (industrial robots) and transportation (autonomous vehicles). The research emphasizes the synergy between model control and machine learning, while collaborating with a wide range of industry partners for autonomous driving research, covering safe and effective planning/control, interactive prediction/decision making, powerful sensing/positioning, and simulation/test/data sets.
About: Graviti
To meet the needs of enterprise AI developers who need high precision, large-scale truth data and effective management of these data sets in the process of turning algorithms into applications, Graviti offers a one-stop set of data set management, data set trading, data annotation services based on SaaS model. An AI data service platform for sandbox model training and model detection and evaluation.
About: AWS(Amazon Cloud)
Amazon Web Services (AWS) is a professional cloud computing service provided by Amazon. Launching in 2006, AWS is the largest cloud computing service provider in the world. One of its key advantages is the ability to replace upfront capital infrastructure costs with lower convertible costs that can be scaled up according to business development.
Click Apply for Java Developer Click Apply for Backend Developer Click Apply for Front-end Developer
