Work

MAIC: Multimodal Active Inference Controller

Year: 2021
Type: dissertation
Abstract: Active inference, a theoretical construct inspired by brain processing, is a promising approach to control artificial agents. Here we present a novel multimodal active inference torque controller for industrial arms that improves the adaptive characteristics of previous active inference approaches but also enables multimodal integration with any other sensor modal- ity (e.g., raw images). We evaluated our model on a 7DoF Franka Emika Panda robot arm and systematically compared its behaviour with previous active inference baselines and classic controllers, analyzing both qualitatively and quantitatively adap- tation capabilities and control accuracy. Results showed improved control accuracy in goal-directed reaching due to the increased representation power, high noise rejection due to multimodal filtering, and adaptability in changes on the environmental conditions and robot parameters without the need to relearn the generative models nor parameter retuning.
Language: English
Open Access status: green
OpenAlex ID: W7020757332
Cites: 22
Cited by:
Related to: 9