Turning Foundation models into “driving mentors”: found-RL enables efficient reinforcement learning for autonomous driving
Peer-Reviewed Publication
Updates every hour. Last Updated: 19-Oct-2026 15:16 ET (19-Oct-2026 19:16 GMT/UTC)
Researchers developed Found-RL, an open-source platform that integrates vision-language models into reinforcement learning for autonomous driving. By using an asynchronous batch inference framework, Found-RL allows large foundation models to provide semantic feedback without slowing down high-frequency simulation training. Experiments in CARLA show that lightweight reinforcement learning agents can achieve competitive driving performance while maintaining real-time policy inference.
A chemical used to make plastic more flexible, once introduced as a safer substitute for older additives now known to be harmful, may itself also be harmful, slowing babies’ growth in the womb, a new study shows.
Reaching net zero sooner would limit long-lasting changes to El Niño and the tropical Pacific, new climate modelling suggests, although heat stored deep in the ocean means changes to the tropical Pacific could persist for centuries.
Elections across the world faced a wide range of pressures on electoral integrity in 2025, from violence and mis- and disinformation to electoral management malfunctions, cyber attacks and natural hazards, according to a new report.
The report, Review of the 2025 Year of Elections: Inclusivity, Transparency and Trust, from the International Institute for Democracy and Electoral Assistance (International IDEA), was written in collaboration with the Electoral Integrity Project and co-authored by Toby James, Professor of Politics & Public Policy at the University of East Anglia (UEA).
It examines 39 countries that held national elections in 2025 and looks at both the risks facing elections and how electoral institutions are responding, with a particular focus on inclusivity, transparency and trust.
“The report shows that all countries, large or small, are facing challenges in ensuring that elections truly empower people. But they also show the value of drawing lessons across borders,” said Prof James, who also co-edited the report.
“Challenges such as the role of money in elections, election disinformation and political polarisation are not confined to single countries. Efforts to improve elections should therefore be global as well.”
Electoral campaigns remain a particularly vulnerable part of the electoral cycle. The report also points to growing pressure on media integrity and campaign finance, as declining press freedom and increasingly digital financial flows create new challenges for electoral processes.