Building Brains for Robots? Natural Intelligence vs Artificial Intelligence

L'Università di Genova con UniWeLab, il laboratorio congiunto UniGe-Webuild, propone una conferenza, in lingua inglese, sul confronto tra intelligenza naturale e intelligenza artificiale.

Il relatore, Rüdiger Dillmann, è direttore di ricerca presso FZI Research Center for Information Technology, e direttore della collana COSMOS (Springer). Dillmann ha fondato l’Istituto di Antropomatica e Robotica presso il Karlsruhe Institute of Technology, è stato coordinatore del Centro di ricerca collaborativo tedesco “Humanoid Robots” e di diversi IP europei su larga scala. Dal 2018 è professore emerito; attualmente fornisce consulenza alle start up e alle PMI dei suoi ex studenti di dottorato.

La partecipazione all'evento è gratuita su prenotazione.

Abstract

The long term goal of this research is to understand, to model and to translate biologic neural principles towards robot control systems. In comparison to conventional computing the brain is superior in terms of energy efficiency, robustness and adaptivity. Thus, we investigate into modeling biologic processes enabling the brain to perform sensomotoric computation and finally to implement it in silicon in form of biomorphic hardware. Todays neuromorphic hardware consists of spiking neural networks (SNNs) which can perform fast and efficient computations with continuous input - output streams based on synaptic plasticity. We focus on brain like senso-motor control principles which are data driven in contrast to model driven AI algorithms. Spiking neural networks have the potential on replicating real neurons representing parts of their biological characteristics. SNNs are capable to perform synaptic spike-based communication with local brain functionalities supporting learning with the help of neural plasticity mechanisms. We assume, that the brain is forming sensor-motor primitives within building blocks composed for object detection, localization, event prediction, and finally the generation and execution of motion and interaction. The combination of neural motion primitives represent complex muscle motor synergies capable to learn complex large scale motions like object recognition, object tracking, target reaching and grasping as well as collision- and obstacle avoidance. Closing the visuomotor loop by mapping the learned visual representation to motor commands show that SNNs learn without any planning algorithms nor inverse kinematics. SNNs are event driven and model free. We introduce deep continuous local learning mechanisms achieving state of the art robot accuracy on event stream benchmarks. Biologically plausible reward-learning rules based on synaptic sampling show that SNNs are capable of learning policies and various movement characteristics. Links between reward-modulated synaptic plasticity and online reinforcement learning show proposing results. An event-driven binocular DVS system is used driven by micro saccades. The spiking feedback information is mapped towards motion generating SNNs driven by reward coupling and prediction error minimization techniques. Future work towards the effective use of neuromorphic vision with emphasis to eye movement, micro saccades, visual affordance learning and high performance event prediction will be discussed. In addition it can be shown, that the brain-inspired computational approach can be extended towards SNN based navigation and mapping forming episodic spatial neural memories with multi-scale learning capabilities.

Conference poster - Dillmann 20 December 2022

Abstract (italian)

Comprendere, modellare e tradurre i principi neurali biologici per i sistemi di controllo dei robot. Rispetto all’informatica convenzionale, il cervello è superiore in termini di efficienza energetica, robustezza e adattabilità. Pertanto, la modellazione dei processi biologici che consentono al cervello di eseguire calcoli senso-motori si può implementarla sotto forma di hardware biomorfico. Ci si concentra sui principi di controllo senso-motorio simili a quelli del cervello, che sono guidati dai dati, in contrasto con gli algoritmi di intelligenza artificiale guidati da modelli. Le reti neurali Spiking hanno il potenziale per replicare i neuroni reali, rappresentando parte delle loro caratteristiche biologiche. L’approccio computazionale ispirato al cervello può essere esteso alla navigazione e alla mappatura basate su SNN, formando memorie neurali spaziali episodiche con capacità di apprendimento multi-scala.

Programma conferenza - Dillmann 20 dicembre 2022