Titulo Estágio
Improving robot adaptability through the evolution of behaviour repertoires
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI-FCTUC
Enquadramento
Autonomous robots are increasingly deployed in dynamic and unpredictable environments where traditional, monolithic control strategies often fall short in terms of adaptability and robustness. An approach for improving their adaptability is to employ Quality-Diversity algorithms to evolve Behaviour Repertoires [1], containing a collection of diverse, simpler and high-performing behaviours, rather than a single and complex one.
Each behaviour in the repertoire is well suited to a specific context, enabling the robot to select, adapt, or combine behaviours in real time based on its own internal state and environmental conditions. This approach improves the robots resilience to novel situations, environmental changes, or partial system failures.
[1] - Cully, A., Clune, J., Tarapore, D., & Mouret, J.-B. (2015). Robots that can adapt like animals. Nature, 521(7553), 503–507. https://doi.org/10.1038/nature14422
Objetivo
The main objective of this dissertation is to investigate how Behaviour Repertoires can be built and used to evolve more adaptable and resilient robots. This will be achieved by focusing on three core aspects:
1 - Construction of a Behaviour Repertoire
Design and implement a Quality-Diversity algorithm capable of generating a diverse set of high-performing behaviours. The repertoire should cover a broad range of situations the robot might encounter, enabling general-purpose adaptability rather than task-specific optimisation.
2 - Design of Meaningful Behavioural Descriptors
Explore and evaluate different strategies for defining behavioural descriptors, i.e., the features that characterise each behaviour within the repertoire. This includes hand-crafted, task-specific descriptors as well as learned alternatives (from the robot's interactions with the environment). The goal is to devise descriptors that enable the creation of a repertoire containing diverse and high performing behaviours for the task at hand.
3 - Efficient Online Controller Selection
Implement and evaluate mechanisms that allow the robot to select or compose behaviours from the repertoire in real time, based on current sensory input or environmental feedback. This includes reactive selection strategies, meta-controllers, or symbolic high-level policies (e.g., evolved via Genetic Programming) that guide the choice of low-level controllers.
The methods devised will be tested in simulation, using PyBullet, comparing their performance to that of a monolythic approach (where a single controller attempts to solve the entire task).
Plano de Trabalhos - Semestre 1
1 - Literature review.
2 - Definition of the techniques and technologies to be used.
3 - Design and development of the initial behaviour repertoire evolution algorithm
4 - Writing of the intermediate report
Plano de Trabalhos - Semestre 2
5 - Exploration of behaviour descriptors
6 - Refinement of the algorithm for evolving behaviour repertoires
7 - Development of the online behaviour selection algorithm
8 - Writing of a scientific article with the main results
9 - Writing of the thesis
Condições
The work will be conducted in the bio-inspired Artificial Intelligence (bAI) group from CISUC.
There is a possibility of the student being awarded a scholarship (Bolsa de Investigação para Licenciado) for at least 6 months, renewable for an equal period by agreement between the advisor and the intern. The scholarship will follow the Fundação para a Ciência e Tecnologia (FCT) monthly stipend guidelines.
Orientador
João Macedo
jmacedo@dei.uc.pt 📩