Autonomous Navigation
[wtm_use_case_cats]
A rover on Mars solely controlled by a team of engineers can only be given instructions to move every 20mins. This is the communication delay between Earth and Mars. Suppose only 5 movement commands can be sent every 20mins, that’s a total of 360 commands in a day. A rover equipped with autonomous navigation capability could make ≥ 5 decisions per min if not per second. It is now limited by the speed of its computer rather than the communication delay.
Benefits for the company
Time and Resource saving
Feasability
Medium
Type of expertise/ AI domain
Convolutional Neural Network – AI
An artificial neural network (ANN) inspired model for satellite navigation is proposed by Mathew C. Wilkinson & Andrew J. Meade. Sequential Function Approximation (SFA) is used to develop an ANN that learns the effect of an engine burn through the variation of its magnitude and direction of burn on the flight path of the spacecraft. The result is compared to the desired state, the subsequent error parameter is used to constantly tweak the neuron parameters. SFA effectively trains its own neuron arrangement through an iterative process of observing the neuron parameter change’s effect on the error parameter. This approach was proven to more computationally efficient than computing dynamics equation or creating a neural map at once.