Embodied Cognitive Systems for Vision-Language- Action Reasoning

• Grant number: 26/07256-0
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: June 01, 2026
• End date: February 29, 2028
• Field of knowledge: Engineering - Electrical Engineering - Industrial Electronics, Electronic Systems and Controls
• Principal Investigator: Flávio Alessandro Serrão Gonçalves
• Grantee: Praveen Kumar
• Host Institution: Instituto de Ciência e Tecnologia. Universidade Estadual Paulista (UNESP). Campus de Sorocaba. Sorocaba , SP, Brazil
• Company: Secretaria de Desenvolvimento Econômico (São Paulo - Estado). Instituto de Pesquisas Tecnológicas S/A (IPT)
• Associated research grant: 20/09850-0 - Applied Artificial Intelligence Research Center: accelerating the evolution of industries toward standard 5.0, AP.PCPE

Abstract

Recent advances in Vision-Language-Action (VLA) models enable AI systems to perceive, reason, and act in complex environments. However, most current approaches remain reactive or task-specific, lacking the cognitive structure required for robust operation in real-world industrial settings marked by uncertainty, partial observability, and continuous interaction with the physical world. This project proposes the investigation of embodied cognitive systems for VLA reasoning in industry, grounded in cognitive systems theory and embodied intelligence. The core hypothesis is that effective industrial VLA systems must tightly integrate perception, language understanding, internal state representation, and action through ongoing interaction with the environment. Rather than treating vision, language, and control as loosely connected modules, this research explores cognitive architectures in which these components are unified to support reasoning, anticipation, and adaptive behavior. Embodiment enables grounding of representations in sensorimotor experience and supports the acquisition of reusable skills. The study focuses on industrial scenarios including collaborative robotics, industrial process configuration, and multisensorial waste management, where agents must interpret multimodal inputs and execute high-level instructions under uncertainty. By positioning VLA models as integral components of broader cognitive systems, this project seeks to advance embodied AI toward intelligent agents that are more robust, interpretable, and industrially relevant. (AU)