Application Platform for Swarm Control

Make it easy to create big control applications for swarms of devices

Why should we care?

We as a community will be creating many Big Control Applications over the next few years and these apps are very different from today’s applications.

The Big Control Applications are different and difficult because they have to cope with real time noisy data, have to generate plans for swarms of mobile devices operating in a dynamic physical environment with very high cost for mistakes.

We don’t want everyone to write his/her application from scratch and do all the heavy lifting.

The APSC makes it easy for developers to create their big control applications for swarms of collaborating devices using abstractions/APIs and a set of building blocks we provide.

Our APSC Approach: Global Map

We maintain a global environment “map” appropriate for a particular swarm of end-devices:

  • a swarm of UAVs might have a semantically labeled obstacle map
  • a swarm of autonomous cars might have a map of traffic congestion
  • a swarm of mobile phones might have a map of user locations

A centralized representation of the state of the environment is critical for situational awareness, which is both the objective as well as the means for achieving this objective in many applications.

The real-time situational awareness can come in the form of three-dimensional models of the environment, sensor measurements, and target tracking and can be used by a variety of applications.

Furthermore, the map can be queried by the end-devices for necessary on-board reactive control.

Our APSC Approach: Common Library Functions

We also provide a set of common library functions that application developers can use:

  • Declarative Planning
  • Entropy Based Exploration
  • Deep Reinforcement Learning
  • Monte-Carlo Tree Search

We are developing deep reinforcement learning architectures that aim to scale to many agents
Publication 1Publication 2

We are studying algorithms for multi-agent active sensing for 3D mapping, and multi-agent information gathering more generally
Publication 1Publication 2

We are developing a fleet of small robotic cars for taking images of an environment for large scale 3D reconstructions

back to top