The emergence of pandemic in early 2020 and the wide spread of heath awareness among the people associated with social distancing and contact-less deliveries of essential items has increased the importance of fully autonomous vehicles more than ever. 

The heavily hit cities of the United States and China have deployed autonomous vehicles for delivering food, medicines, essential items, and other goods during this Corona virus out-break. 

The AV manufacturers and the technology players are strategically moving every step towards the viability of AVs. The industry has understood that the advancement and improvement in the technology should not be kept at halt. Unlike past years, the companies are spending the share of the revenue pie on R&D in a very sustainable manner. This has resulted the delay in the launch timelines of more than few AV companies. Also, the highly autonomous vehicles are still in their prototype and testing phase and have not yet reached the stage of production and deployment. Companies are rigorously testing and validating their AVs, control units, components, algorithms, and software to speed-up the journey. However, there are still many challenges that needs to be faced and surmounted. 

 This article is intended to address one among the several challenges in the autonomous vehicle market and that’s “Annotation or the Training Data”. 

 What drives a vehicle autonomously is heavily dependent on the capability of the vehicle to precisely spot any object, any signs, or markings around its radius of min 5 meters. Along with this, identification of the movements of the objects (including vehicles, pedestrians, cyclists, and animals) and foreseeing their path is another most important element in a self-driving vehicle. 

However, this detection, tracking, and identification is nothing but just a collection of information and knowledge about the nearby environment. Unless the data is actionable, and the vehicle is made intelligent enough to decide its path of driving, the vehicle will not be considered safe for driving without a human supervision. 

For this, the collected data needs to be processed in a perception model to extract relevant information from the immediate surrounding, allowing the vehicle to decide its course of action. This is most essential for the safe operation of AVs. 

 Machine learning is the foundation for this perception model. The trained developers are striving to build an effective and robust perception algorithm that can help AVs drive safely on public roads. 

These algorithms must undergo rigorous testing, validation, and verification with multiple sets of scenarios. However, to make the ML algorithm work, it needs a consistent high-quality training data that is responsible in determining the behavior of the product. And lack of this is the major challenge.

 Annotell is one of the high potential companies in the space of autonomous driving aiming at solving the inadequacy of the high-quality annotated data for supervised machine learning. 

This Sweden based company was founded by Oscar Petersson and Daniel Langkilde in 2018. They then realized the issues of the ground-truth data labelling and the in-consistency of training data sets that machine learning algorithm developers are dealing with for mission critical applications like automotive and transportation. 

So, Annotell was established with an aim to offer a unique platform that will allow the AV technology companies to offload their burden of ground truth labelling and infrastructure management and focus completely on the development of autonomous solutions. 

 The company is following a designed systematic approach to feed the ML model with training data that meets required level of details and consistency. The platform involves two major steps- the first is setting annotation guidelines and the second is annotation engine. 

 In the first step, Annotell addresses the major often overlooked gap of creating proper guideline that complies to the client’s expected output. These expectations on annotations are set-up by identifying two key dimension- the required range of acceptable solutions and the required pixel accuracy or number of properties. 

These factors are expressed as an annotation guideline that annotators can follow ensuring the coverage of most important cases and reducing the time and overhead of managing the project. In second step, the annotation engine effectively creates the large amount of consistent annotations. 

The tool offers flexible level of automation that is capable of avoiding wastage of time in reviewing and correcting bad annotations, subsequently saving the project cost. 

 So, with scalable platform that offers high level of machine assisted labelling, coupled with 100% automotive industry expertise and a defined approach towards annotation, Annotell is trying to solve the current issue of lack of consistent training data for testing and validation of autonomous vehicle. 

The company has gained many recognized clients and partners since its inception and pursuing its growth plans at full throttle. Zenuity, Volvo Group, Scania, CEVT, Tobii Eyetracking, Smart Eye, and mobilityXlab are few among the list of its customers and partners.