Tiny(43mb) voice-first decision model for robotics (huggingface.co)

NeuDecide has introduced a groundbreaking voice-action decision model specifically designed for robotics, featuring an impressively compact size of just 43 MB. This model directly converts spoken commands into actionable function calls without requiring a transcript, allowing it to operate independently of GPUs or internet connections once downloaded. Users can define the model's capabilities through simple JSON schemas at inference time, making it incredibly flexible. For example, a command like "clean the bathroom" can prompt a robot vacuum to execute that task seamlessly. This innovation is significant for the AI/ML community as it sets a new standard for lightweight, efficient decision-making models in robotics. NeuDecide leverages a pre-trained streaming English speech encoder combined with a tool-calling encoder-decoder system, achieving impressive tool accuracy rates of 75.5% and outperforming larger models in exact match metrics. Its architecture allows it to run on various devices, from MacBooks to Raspberry Pis, with minimal latency. Additionally, the model is user-friendly, requiring only a few lines of code to implement, making it accessible for developers looking to integrate voice control in robotics without extensive computational overhead.