Leveraging Large Model-Based Embodied Intelligence to Enhance Financial Service Robots

Multimodal large models-based embodied intelligence, which integrates cognitive and reasoning abilities with interaction in the physical world, allows agents to perceive, plan, and act in real world. In the financial services industry, where human interaction and data management are critical, embodied intelligence offers new opportunities and challenges. We have exploited multimodal model-based embodied intelligence to enhance branch service robots in the financial industry and optimize the assessment of insurance claims. These implementations were powered by training multimodal models on NVIDIA data center GPUs. With the NeMo, we achieved an acceleration of 20-25% in model training. ISAAC and Omniverse tools enabled us to perform virtual environment simulations and modeling, while TensorRT provided inference acceleration for deployment on Jetson platforms. Our practices have been employed in the financial industry, improving customer Net Promoter Scores and reducing operational costs by approximately 10%.