Addverb is expanding its robotics portfolio as it works towards integrating Physical AI into industrial and warehouse automation, developing its own Vision-Language-Action (VLA) model for warehouse applications while adapting advanced open models for specific robotic tasks, with efforts spanning robotic perception, navigation, decision-making, manipulation and the integration of AI with physical systems. In an interview, Sangeet Kumar, CEO and Co-founder of Addverb, discussed the company’s approach to proprietary AI development, the technical and commercial progress of its Elixis-W humanoid platform, its revenue target for FY2026–27 and the role of international markets. Kumar said Addverb has been investing in Physical AI for over three years and generates about 1,000 hours of robot data every week to train and improve its systems for real-world environments. The proprietary VLA model processes inputs from multiple cameras, interprets instructions and translates them into robot movements, with training combining visual reasoning, knowledge distillation from a larger AI model and motion data captured through robot demonstrations. Addverb is working with advanced open models such as π0.7 from Physical Intelligence and Cosmos 3 from NVIDIA, adapting them to warehouse-specific applications; work with π0.7 includes bolt packing and Ethernet cable insertion, while with Cosmos 3 the company is exploring bolt pick-up and box packing. Addverb has secured commercial orders for Elixis-W, which is designed for industrial applications.