Heterogeneous Robot Planning using Active SLAM


Project Member: Alexander Schperberg (collaboration is welcome, if interested, feel free to contact me).


This project will build a framework to enable heterogenous (legged, driving, and/or flying) robots to achieve some cooperative goal while considering uncertainty from exteroceptive and proprioceptive sensor measurements. The high-level objective is to address the fundamental problem of Active SLAM, or sometimes referred to as simultaneous planning, localization, and mapping. To accomplish this, requires integration of multiple algorithms and components. The following three areas will be used to build our multi-robot planning framework:

  1. Optimization (employ path planning algorithms such as model predictive control or rapidly exploring random trees while accounting for complex robot dynamics)
  2. Observation (use state of the art visual-inertial algorithms for mapping and localization)
  3. Behavior (implement reinforcement learning to discover optimal behavior of multiple agents to achieve a cooperative target goal)

reference for picture: Ravankar, A.; Ravankar, A.A.; Kobayashi, Y.; Emaru, T. Hitchhiking Robots: A Collaborative Approach for Efficient Multi-Robot Navigation in Indoor Environments. Sensors 2017, 17, 1878.

An important aspect of this project will include building dense 3D maps (as seen in the image above), which facilitate collision avoidance for the planner and provide information on environmental uncertainty.

After building this framework, we will apply it on a diverse set of robotic systems, including drones, Roombas, and even a quadrupedal robot. The project is considered successful if a heterogenous team of robots can behave in such a way that will avoid collisions, achieve some target goal, and ensure that uncertainty in localization is below a specified threshold.