AI Autopilot Final Year Project: Automation and intelligent - - PowerPoint PPT Presentation

AI Autopilot

The Data Analysis Bureau

Final Year Project: Automation and intelligent

• ptimisation in high performance sailing boats

Project Introduction

• Machine Learning approaches to the Helmsman problem

– Controlling the rudder input based on recorded data

• Jack Trigger data collected from onboard sensor- and

processing system (NKE systems)

• Various data formats and additional data sources
• A problem split into the domains of:

– Control theory – Data science – Applied machine learning – Eventually IoT

[1] Jack Trigger Racing, www.triggerracing.com

Background and previous work

Background and motivation:

• Recent explosion in the development of ML for the

purpose of autonomous driving and other applications

• A desire to transfer recent developments in methodology

and toolboxes to the realm of high performance sailing

• Within high performance single-handed sailing:

The desire to emulate a human driver and outperform existing, traditional autopilots

Skype conversation with Dr. Pieter Adriaans

Data Sources

Processor HR BoxWifi

Sensor Array

Topline Bus

Adrena Export: .csv NKE Proprietary: .NKZ NMEA-0183: .LOG (txt)

Data Sources

.csv .NKZ => .csv

• 17 files
• 41 features
• ~200 hours
• 1 Hz
• No rudder data
• No pilot data
• 32 files
• 188 (~50) features
• 16 hours
• 25 Hz
• Rudder data
• Pilot data

Aim: Alleviate the workload of a solo sailor and

• ptimise Velocity Made Good (VMG),

rather than just keeping course

Two different approaches to the same problem:

• Reinforcement Learning
• Supervised Learning

Reinforcement Learning

Agent Environment action state reward Boat Wind Sea

Rudder angle

Rate of convergence to the desired location Start Goal

• Velocity
• Orientation

Database

• Wind
• Sea State
• Location

Agent (Actor) Dynamic Model of the Boat (Critic) Rudder Angle BOAT (model) VMG Sea State Input ENVIRONMENT (real) State

LSTM – Long-Short-Term Memory Neural Networks

LSTM consist of memory blocks called cells. Each cell has three gates:

• 1. Input gate – which information is

useful at this particular level

• 2. Forget gate – which information

is no longer relevant and could be forgotten.

• 3. Output gate – what information

is relevant for the next cell

Allows for long-term memory, so history of boat dynamics can be included in the estimation of the next state.

. . .

Wind Boat Speed Current Boat Orientation Boat Position LSTM

... t-3 t-2 t-1 t

. . .

Control Action

. . .

Boat Speed Boat Orientation Boat Position

t+1

• Velocity
• Orientation

Database

• Wind
• Sea State
• Location

Agent (Actor) Dynamic Model of the Boat (Critic) Rudder Angle BOAT (model) VMG Sea State Input ENVIRONMENT (real) State

Database

• Wind
• Sea State
• Location

Sea State Input ENVIRONMENT (real) LSTM Rudder Angle

. . .

Wind Boat Speed Current Boat Orientation Boat Position LSTM

... t-3 t-2 t-1 t

. . .

Control Action Control Action

t+1

Rudder Angle

Moving to the Cloud

Future: Edge Computing