Autonomous Agents Networks (AAN) is a Linux Foundation Edge ("LF Edge") Akraino blueprint under the Physical AI blueprint family. A coming onslaught of autonomous machines, including mobile robots, drones, self-driving vehicles, and many more are being deployed in different contexts and in close proximity to human activities. There is an urgent need for architectural frameworks and reference as well as standards for how these “physical agents” will interact with each other and with humans in safe, reliable, and collaborative ways. The AAN blueprint shows how agents -- which are differentiated from machines by their ability to cooperate, achieve goals, and operate autonmously -- will utilize both AI multimodal models and communications technologies and network standards such as P2P (peer-to-peer), cellular, and satellite to 1) collaborate without cloud dependencies and 2) notify each other and humans of important safety related events they detect and observe.
On-device multimodal small models and P2P communications allow machines in manufacturing, construction, agriculture, and other machine-intensive industrial sectors to communicate directly, regardless of cloud connectivity issues or WiFi impediments such as metal or other interference, distance, landscape obstructions, etc. Direct communications, both multicast and bi-directional, are supported by cellular and satcom networks. This approach is both commercially viable and maintains a high priority on safety
Cameras, water level sensors, drones, meteorological stations and dam leak systems exchange data, send alerts, monitor overall systems and feed different datasets concerning rain level, floods (local and regional flow) etc. Agents include audio and video communication to allow real-time human interaction during rescue operations
Mobile robots, drones, and vehicles detect, identify, and report incidents to 1) other machines within a 1 to 5 mile radius, and 2) First Responders depending on incident priority level. Machines from different providers and vendors must observe and understand their surroundings, interact with humans if necessary, and share information about safety - human safety as well as property and other machines. P2P cell and satcom network protocols eliminate dependencies on cloud communications subject to vendor incompatibility / lack of cooperation and unreliable connectivity depending on environmental conditions
On-device multimodal small model app for human safety in close proximity to robots and automated vehicles. Personal apps can utilize visual and audio input to monitor users' surroundings for machine conditions and behavior, for example when riding as a passenger in a robotaxi, users can attach a smartphone USB camera to a headrest or other elevated location and warn on potentially dangerous conditions such as imminent pedestrians or pets. As another example factory workers can wear their phone as a body camera to monitor automated machines (eg. a forklift) for human proximity or other behavior
