TeNet: Text-to-Network for Fast and Energy-Efficient Robot Control
Large Language Models, such as ChatGPT, are more and more often applied to equip robots with the ability to receive instructions from humans to solve tasks and problems. Enabling robots to understand natural language helps us to lower the barrier and enable people without special training to program robots and deploy them easier in the real world. However, large language models come at a cost: They are large, with billions of parameters, requiring special and expensive computers, consume a lot of energy, and need several seconds to process data and produce outputs. This makes it tricky to deploy them directly on robots, as they often cannot carry large payloads, such as large computers, and if, for example, mobile robots draw energy from batteries we want to optimize their energy consumption such that they can function as long as possible without the need to recharge and pause their task.
In this project Kevin Luck will see if he can solve with his colleagues some of these problems by training large language models to produce smaller networks, which can be queried faster, need less energy and can be deployed directly on a robot. With this he hopes to not only enable robots to be able to solve tasks which require fast real-time coordination and short reaction times, like throwing, balancing, object-manipulation etc., but also decrease the amount of energy needed for running these language models hence contribution to a more sustainable approach for robotics and real world applications.
About this call
This call is geared towards ideas and initiatives which tackle one or more of the challenges of the national AI research agenda AIREA-NL and collaborate with one or more European collaborative partner organization based outside of the Netherlands. The proposed research is ground‐breaking and involves a real risk of failure. What counts is that all results, be they positive or negative, must contribute to the advancement of science. This is “blue‐sky thinking”. All proposals are anonymous, to ensure that the assessments are based purely on the research ideas they contain.
About the National Growth Fund programme AiNed
This Call for proposals is part of the National Growth Fund programme AiNed. This programme promotes the development and application of artificial intelligence (AI) in Dutch businesses and governments, and was developed by the Netherlands AI Coalition. Bringing the Netherlands into the leading group of AI countries for prosperity and well-being. That is the joint goal of the Dutch AI Coalition and the AiNed programme. Through a collaboration of public and private parties taking essential steps in developing and applying AI in various sectors and for important economic and societal challenges. In 2021, the AiNed proposal received a grant of € 204.5 million from the National Growth Fund.
National Growth Fund programmes
NWO runs thematic programmes for research, knowledge development and innovation funded by the National Growth Fund. The results are used in innovations and organisations, thus contributing to the sustainable earning capacity and broad welfare of the Netherlands. The programmes bring together parties from the entire knowledge chain, both public and private.
Read more about the other grant winners on the website of NWO