MULTI-SENSOR FUSION AND DEEP EVOLUTIONARY OPTIMISATION: IMPROVED EMBODIED COGNITION AND MANIPULATION ANALYSIS IN HUMANOID ROBOTS
Hayder M. Ali, S. Bairavel, L. Arokia Jesu Prabhu, Anusha Papasani, A. N. V. K. Swarupa, T. Thirumalaiakumari, Bharathi Ramesh Kumar, Sudhakar Sengan
Abstract: Humanoid Robots (HRs) face challenges integrating sensorimotor data due to inadequate sensor fusion methods and suboptimal control models. The research results of the current study should inform the implementation of a computational model that optimises and enhances the success rate of embodied cognition and information manipulation. The machine learning algorithm incorporates Hierarchical Multi-Sensor Fusion (MSF) and Deep Evolutionary Optimization (DEO), both introduced as computational methods into the learning process. Using attention-weighted neural network (NN) models, the machine learning (ML) algorithm integrates visual, tactile, proprioceptive, and force feedback information throughout its model. To achieve the objective of the evolution process, artificial intelligence NNs evolve using Genetic Algorithms (GAs) that integrate multi-objective fitness functions. By implementing focused algorithms specifically designed to build a cognitive network (CN), the evolutionary algorithm can optimize network topologies and NN parameter settings. Empirical validation was conducted on an HR with 22° of freedom of movement. When the results were compared with baseline methods, the phenomenon was found to be a 34.7% increase in manipulation success rates, a 28.3% reduction in task completion time (TCT), and a 41.2% increase in adaptation speed (AS). The theoretical model also identified emergent cognitive functions. These practices included adaptive body schema learning, predictive motor control, and context-aware attention allocation.
Keywords: Embodied cognition; Humanoid robotics; adaptive manipulation; Multi-sensor fusion; Cognitive networks; Deep evolutionary optimization
DOI: 10.24874/PES08.03.014
Recieved: 12.05.2026 Revised: 30.06.2026 Accepted: 03.08.2026
UDC:
Reads: 11 
