SBIR Phase I: Development of a Natural Language Dialogue System for the Blind and Visually Impaired to Enable Greater Efficiency in Remote Assistance

这个小企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是大大提高美国近2200万盲人或视力受损者的生活质量和经济独立性。经济效益包括显著减少因视力障碍导致的生产力下降造成的近1000亿美元的年度经济损失，以及盲人或视力障碍者的年度社会服务成本。 该项目将创造低成本、高价值的工具和服务，使盲人和视力受损者获得与正常人相同的环境意识。在规模上，这项技术将直接雇用超过20，000人来支持这项服务，同时也为数百万盲人/视障人士提供加入劳动力大军的工具和机会。拟议的项目将为盲人开发一种实时的、改善生活的服务，通过机器学习和远程人类助理的组合来创建一个实时的、半虚拟的个人助理，具有面对面的、训练有素的人类助理的质量。 在拟议的项目中，核心创新是一个高级机器智能工具，该工具是通过集成最先进的自然语言理解（NLU）软件和图像识别与分析（IA）软件以及新颖的代理间路由和最先进的技术创建的。最先进的优雅降级，或者由软件或人工代理进行无缝错误处理和解决。 尽管NLU和IA代理取得了令人难以置信的进步，但将多个不同的软件代理集成到一个特定于上下文的应用程序中仍然是一项困难且有风险的开发工作，值得NSF资助。通过这样做，我们将（1）通过更丰富的环境反馈来增强用户体验，（2）提高我们人类代理的生产力，允许单个人类代理为更多用户提供服务，从而创造可扩展性，使服务能够自我维持和负担得起，最后（3）增加用户？通过与机器工具的互动，而不是与人类代理人的情感和社会互动，体验个人独立性。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to dramatically increase the quality of life and economic independence for the nearly 22 million blind or visually impaired people in the US. The economic benefits include a significant reduction in the nearly $100B in annual economic losses from lower productivity due to visual impairment and in annual cost of social services to the blind or visually impaired. This project will create low-cost, high-value tools and services that will give blind and visually impaired people the same level of environmental awareness as fully sighted people. At scale, this technology will result in the direct employment of more than 20,000 people to support the service, while also providing millions blind/visually impaired people with the tools and opportunity to join the work force. The proposed project will develop a real-time, life-enhancing service to the blind enabled by a mix of machine learning and a remote human assistant to create a real-time, semi-virtual, personal assistant with the quality of an in-person, trained human assistant. In the proposed project, the core innovation is a high-level machine intelligence tool created by integration of state-of-the-art Natural Language Understanding (NLU) software and Image Recognition and Analysis (IA) software with novel inter-agent routing and state-of-the-art Graceful Degradation, or seamless error handling and resolution by either software or human agents. Despite the incredible advancements in NLU and IA agents, integration of multiple different software agents into a single, context-specific application remains a difficult and risky development effort worthy of funding by the NSF. By doing this, we will (1) enhance the user experience with richer environmental feedback, (2) enhance the productivity of our human agents, allowing individual human agents to serve many more users and thereby create scalability that will make the service both self-sustaining and affordable, and finally (3) increase the user?s experience of personal independence through interaction with machine-based tools rather than emotionally and socially charged interactions with human agents.