Some of the new products on show
Software, tools and hardware for L2 to L5
Reliable neural networks that form the heart of the newest generation of perception algorithms need large, diverse, contextually rich annotated data sets for training and verification. Therefore, data will determine the leaders in the automated driving race – those who want to win need to have the most efficient data pipeline in place. One should consider several aspects: cost of operating the pipeline, pipeline throughput, diversity of collected data, selective smart collection (based on iterative benchmarking), quality of labeled data, reusability of data, etc.
Using synthetic data for training automated driving software is one of the hottest topics in the industry, but there are still many questions marks. How can one ensure that synthetic data is representative of the real world for training or validation? What is the right mix of real and artificial data?
On the other hand, as using solely artificial data seems unfeasible for the near future, efficient and affordable collection and annotation of real-world data are also crucial. Smart, scalable automation is the key here. How can one make sure that one is gathering the right miles? How can one annotate dynamic and static data with great efficiency?
Virtual validation is another inevitable step to reach fully automated driving. Similar to training data collection, driving millions of ‘boring’ miles on the road might tell you about the robustness of your system. Still, it will definitely not cover the close-to-infinite problem space your autopilot is facing. The answer is to create a deterministic and high-fidelity simulated environment that allows thousands of variations of the same scenario to be tested in different weather conditions, different locations, with various traffic participants. It is the only way to force your self-driving system to its limits, into corner cases and potentially even dangerous situations. Many challenges lie ahead as the industry strives for ever higher levels of automation, and the availability of affordable data is definitely one of them.
aiMotive is one of the largest independent automotive technology powerhouses working on automated driving solutions from L2 to L5. The company delivers an integrated portfolio of software, tools and hardware products complemented by proprietary data management tools, enabling customers to rapidly develop and deploy production automated driving features that combine in-house expertise with aiMotive modular capabilities while achieving substantial reductions in development costs and timescales. The company’s product portfolio has been validated in mass-production programs. Its lightweight execution stack and sensor-agnostic, reusable data pipeline accelerate customers’ time-to-market.