Spatial AI Agents
The Active Occupiers of Space
Occupants of space again consume and traverse the opportunity afforded by spatial environments.
For example:
- People, robots, walls, and furniture can take up location.
- A market can become saturated by competitors, making entry by new startups difficult or untenably expensive.
- Governments can deploy sanctions making it illegal to do business with particular individuals and countries.
- Your calendar can be so full it can’t accommodate another appointment.
There are two kinds of occupants:
- agents, who both consume but also proactively traverse spatial opportunity, like moving from one location to another, and
- obstacles, which constrain agent behavior by consuming opportunity, temporarily or permanently.
Agents can sense various properties of the environment (e.g. light or sound), interact with the environment (pushing a door open), and reason about how to traverse the environment effectively for some purpose or goal. In particular, they can leverage:
- data they sense (e.g., a sign that says “One way”),
- spatial memory of what they’ve experienced already (e.g., having been in a specific environment before), and
- anything they can infer (e.g., Restrooms are usually near the bar in a restaurant.).
As well as properties of the environment, agents may also detect and respond to the presence of other agents. If we think about walking through a crowded transit station, we predict where our neighbors are walking so we can avoid colliding with them. To ease our passage, we may adjust our path to join the flow of travelers headed in the same direction. We may run up an empty staircase instead of waiting for a long line at the escalator. Generalizing this notion may mean considering the positions of other candidates in a political environment or robots coordinating path-finding in a warehouse.
And there are established and emerging AI techniques for each of these abilities. Sensing can be augmented by new hardware or computer vision algorithms. Spatial memory can be enhanced through storing and analyzing experiences, like a robot dog recording LiDAR data for future use. Inference ability can be enhanced by machine learning.
All of this is in service of an agent updating its model of the world, its model of how it can predict the changes in opportunity ahead and thus make effective decisions.