Ever dreamt of a car that not only drives itself but also explains its actions?
Meet LINGO-1, the brainchild of the innovators at Wayve, designed to revolutionise autonomous driving using the power of language.
Large Language Models (LLMs) have been making waves in the AI domain, automating numerous tasks with their human-like responses to natural language queries. But Wayve is pushing the envelope further with Vision-Language-Action Models (VLAMs) that combine images, driving data, and language.
The integration of language in training autonomous driving models is a relatively new concept. However, it holds the potential to transform how we interpret, explain, and train our driving models. Imagine a world where you can ask your autonomous vehicle why it took a particular action, and it responds in a language you understand. It's not a distant dream anymore; it's the reality that VLAMs promise.
Introducing LINGO-1, a driving commentator that uses language to explain the reasoning behind driving actions. Trained on various vision and language data sources, LINGO-1 can perform visual question answering (VQA) tasks such as perception, counterfactuals, planning, reasoning, and attention.
However, like all pioneering technologies, LINGO-1 has its limitations. Currently, it's trained on Central London driving experiences and internet-scale text, limiting its generalisation capabilities. Plus, it suffers from hallucinations, a well-known problem in large language models.
Despite these challenges, the Wayve team is hard at work to refine LINGO-1. They're exploring the use of natural language to build foundation models for end-to-end autonomous driving. As we push the boundaries of embodied AI, vision-language-action models like LINGO-1 could have a significant impact.
So, fasten your seat belts as we drive into the future of autonomous driving with Wayve's LINGO-1. It's not just about getting from point A to B anymore; it's about understanding the journey.
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comments