Advanced Air Mobility

Multi Agent Path Planning, MPC, Deep RL, MuZero, Pseudospectral Methods

AAM/UAM air traffic management problems were formulated and solved via various optimization approaches.

Model Predictive Control Approach

We solved high-capacity urban air traffic optimization problems using model predictive control with constraint aggregation techniques. In a multi-agent environment, agent trajectories were optimized based on origin-destination pairs for the following optimization paradigms:

  • Centralized optimization at each time step
  • Sequential optimization at each time step
  • Sequential optimization for the entire flight trajectory
AAM/UAM Air Traffic Optimization Using Model Predictive Control (MPC) and Constraint Aggregation
The multi-agent path planning problem is being solved at each time instant using centralized MPC framework)
Optimal Paths at Various Time Steps (The planning problem is initialized in a symmetric manner and solved using centralized MPC framework)

Deep Reinforcement Learning Approach

Inspired by the success of DeepMind’s deep reinforcement learning algorithm, MuZero, in mastering games without any knowledge about the rules and the environment, we investigated the feasibility of using MuZero for AAM research by casting the dynamic obstacle avoidance problem as a game.

Dynamic Obstacle Avoidance Achieved by Path Planning Networks Trained Using the MuZero Algorithm

Mixed Integer Linear Programming

In this project led by Kotwicz Herniczek, corridor design problem were formulated as a minimum cost multi-commodity flow problem. The goal is to generate a corridor network that is globally optimum in terms of trip length, airspace complexity, and corridor network size using linear programming (LP) and mixed integer linear programming (MILP) problems and Gurobi.

Pseudospectral Methods

Pseudospectral methods were explored to achieve minimum time multi agent paths using using OpenMDAO, Dymos, and SNOPT.

Contributed to Extensible Trajectory Optimization Library (ETOL) for Dymos integration

References