As human behavior depends on a tightly controlled perception-action cycle that carefully considers uncertainty and risk, so should the behavior of an autonomous agent. But this is not trivial to attain. How much risk should an autonomous agent take? And how does it determine that risk?
Can I cross? Is the cyclist seeing me? Will that car actually stop? A familiar train of thoughts for a pedestrian navigating an urban environment. How do we ensure that autonomous vehicles make life easier, not more complex?
The ACT project bridges Neuroscience, Behavioral Psychology, Engineering, Robotics and AI to study interactions with humans and autonomous systems and develop new applications for safe navigation
The ACT project organizes various outreach activities, including co-creation workshops for governmental and societal partners; technology demonstrations and public school-based awareness programs. Events will be announced here.
The ACT project helps safeguard society from entering a suboptimal, risky and irreversible transition to living with autonomous vehicles. It does so by developing AI knowledge whereby society is in the lead and putting the human central in AI technology.
The ACT project aims to 1) attain Increase insight into the modes of human behavior under uncertainty 2) develop algorithms that capture behavior under uncertainty and 3) establish measures and guidelines that determine user friendly behavior
Recent technologies like GPS, smartphones, smart wearables, drones, or self-driving vehicles are changing the way we interact with others in our environment. While smart phones have already become ubiquitous and indispensable, others technologies, like drones and self-driving cars, are much more difficult to integrate, also for society at large. Developing self-driving cars behave annoyingly unpredictable for other drivers, and can fail dramatically in complex city environments, and similar problems exist with drones. How do we make the behavior of autonomous vehicles predictable for humans, to avoid conflict and dangerous situations for other traffic participants? Humans are very good at predicting and adequately responding to behaviors of other humans, even in demanding environments; machines lack this human-predictable behavior needed to make urban mobility safe. This project will chart out these human behaviors and implement them as “neuroware” to artificial intelligence systems that can guide autonomous vehicles, like drones and self-driving cars, in a safe way. Scientific, industrial, and societal partners capitalize on a combination of psychology, neuroscience, artificial intelligence, engineering and robotics to develop integrated solutions for the interactions between humans and autonomous systems, and design protocols, rules and innovative systems for urban mobility and safety.
The consortium encompasses a large range of expertise in experimental and computational neuroscience (in particular, brain mechanisms behind multisensory perception and sensorimotor control), Psychology, AI and robotics (including Bayesian inference, machine learning, deep learning, bio-inspired computation, motion planning and control), and human-machine interactions (behavioral analytics, machine learning, engineering, control theory). Industrial partners bring both hardware and software engineering expertise as well as platforms of what? (2getthere, IMEC, NXP, AIIM). Societal partners align the outcomes with future guidelines and socially desirable outcomes of what? through their expertise in these fields (VVN, RDW, SWOV).
The ACT project encompasses three main project drives, integrating into a set of pre-defined use cases.
prof dr Pieter Medendorp 4 PhD's & Postdocs
Goals: Build-up a fundamental understanding of how human brains deal with uncertainty in the real world. Translate these insights into frameworks that can be applied in autonomous systems. Demonstrate uncertainty-optimized autonomous agents. This sub-project involves as PIs: prof dr Pieter Medendorp (RU), dr Jorge Meijas (UvA), prof dr Cyriel Pennartz (UvA), and prof dr Sander Bohte (CWI, UvA).
dr Javier Alonso Mora 4 PhD's & Postdocs
Goals: Create a fundamental understanding of how autonomous agents can cope with uncertainty. Provide means for computing performance guarantees of autonomous AI systems under uncertainty. Demonstrate risk-aware autonomous agents that are demonstrably trustable and predictable. This sub-project involves as PIs: dr Javier Alsono Mora (TUD), dr Jens Kober (TUD), prof dr Robert Babuska (TUD), prof dr Guido de Croon (TUD), and prof dr Sander Bohte (CWI, UvA).
prof dr Marieke Martens 3 PhD's & Postdocs
Goals: Understand how autonomous agents can cope with human uncertainty. Develop human-predictable perception, planning, and decision-making in interactive environments. Deliver guidelines and demonstration of human-aware behavior and planning of autonomous agents. This sub-project involves as PIs: prof dr Marieke Martens (TU/e and TNO) and dr Bastiaan Petermeijer (NLR).
prof dr Sander Bohte 2 PhD's & Postdocs
Use cases will (1) focus on efficient multi-sensory information processing with noisy and constrained resources, while maintaining safety guarantees; (2) focus on operator-drone coordination, where efficiency, robustness, and reliability are coupled to vehicle and mission specific guidelines in a risk-sensitive way, to achieve human-predictable and human-trustable behavior; and (3) focus on autonomous driving in a typical busy and cluttered Dutch setting. All partners in the project are involved with the use-cases.
The ACT project is guided by the Steering Board.
Senior researcher at CWI and part-time professor of Computational Neuroscience at UvA and part-time professor of bio-inspired artificial neural networks at RUG.
Professor of Sensorimotor Neuroscience, Donders Institute, Radboud University Nijmegen.
Associate professor at TU Delft.
Director of Science at TNO Traffic & Transport; professor of Human-Machine Integration at TUe.
Noteworthy developments in the project and related news.
With most people having joined the project, we had our first in-person kick-off meeting at Hotel De Werelt in Lunteren.
Almost a year has passed since our project kick-off, and the next yearly general meeting is coming into view: on March 15th we are expecting all of you in Nijmegen to enlighten all of us on your progress and to discuss how we can make the most out of the project together.
The next general meeting is held on November 29th, 2023 at the 3ME faculty of TUD.
The ACT project comprises of a talented and diverse team of researchers, with the heart of the research being carried out by the PhD-students and Postdocs, including:
Working at TUD on (self-supervised/online/meta) learning robust and efficient neuromorphic perception and processing for vision-based autonomous drone navigation.
Jesse Hasenaars
PhD-studentNeuromorphic Uncertainty Estimation at CWI.
Tao Sun
PostdocProbabilistic intention prediction of traffic participants in urban environments at TUD.
Anna Mészáros
PhD-studentPlanning in uncertain environments, at TUD
Khaled Mustafa
PhD-studentIntuitive Human-Machine interfacing at NLR.
Nischal Lingam
PhD-studentCopyright © All rights reserved.