Melbourne Robotics Prototyping

Our CTO (Luke Cole) previously worked for Hemisphere GPS (orginally called BEELINE, and now bought out by AgJunction) as a "Robotics Engineer" implementing auto-guidance solutions for various quadbikes and agriculture tractors that was used by 100's of vehicles around the world.

For 10 years, starting as a teenager in 1998 - Luke Cole has also worked for leading research institutes and companies such as NICTA (now called CSIRO Data61), CSIRO, Seeing Machines and ANU Robotics System Lab (lead by Alex Zelinsky, who received a rare prestigious AO award in 2017 and was Defence Scientist of Australia from 2012 for 6 years). Luke's worked included various autonomous mobile robot projects, involving computer vision, and even a self-driving car early 2000's. Back then OpenCV and ROS didn't exist, so we did a "roll-your-own" called VisLib and DROS comprised of 364,578 lines of code.

Lance Cole has also worked at NICTA and has a background of various hardware development, such as working for a contract company to the US millary (EOS), building the Common Remotely Operated Weapon Station (CROWS).

We have a long-standing robotics experience - our engineers offering Melbourne Robotics Prototyping, robot development and robotics custom software services have something like a combined 50 years worth of experience in the robotics field, from teleoperated and semi-autonomous mobile robotic applications, to custom software and/or custom hardware for general automation solutions, signal processing and control systems. Our knowledge base started in NSW and ACT, but now we primarily service East Coast Queensland.

We have developed autonomous mobile robots for air, underwater and ground. Our professional experience started with developing various control and sensor systems for small underwater vehicles in the late 1990's. We where fortunate enough to have been involved in one of the first self-driving car R&D projects back in early 2000's using a 4WD (to support computers for the large processing requirements). For an overseas client we developed a low profile (70mm high) semi-autonomous mobile robot platform for manikin/dummy mounting to simulate people moving (for vehicle crash safety and collision development by German R&D car manufactures). We have been fortunate enough to have been invited to the bulk of the German R&D car manufactures where they develop and test self-driving and driver assist development systems. We have developed various solutions for 2cm accuracy precision steering-guidance solutions for various types of Agriculture tractors (via the CAN bus and ad-hoc methods), which are still used by 1000's of tractors all over the world. We have retro fitted Quadbikes to allow semi-autonomous control via GPS way points. We have custom developed various indoor mobile robotics for indoor localistion and SLAM R&D purposes.

Robot navigation is the task where an autonomous robot moves safely from one location to another. This involves three primary questions:

1. “Where am I?” which is known as robotic localisation (hard).
2. “Where am I going?” which is known as goal recognition (typically provided by a human).
3. “How do I get there without collisions?” - path planning (easy) and obstacle avoidance (hard).

For robotic localisation and obstacle avoidance we use sensors to solve the problem. To move along the planned path, we use control systems.

We have a deep understanding of signal processing and sensors of various types. We appreciate sensing is a hard problem. There is no one-size-fits all solution. Odemetry (wheel encoders) provide a cost-effective method to measure relative position. however suffer from wheel slip and errors are accumlate over time. GPS only works outdoors, effected by trees/buildings, and without a nearby basestation (for expensive DGPS/RTK) the absolute position error is several meters. IMU (accelerometers + gyros + Magnetometer) suffer from drift errors and noise error causing ``random walk'' when integrated. Magnetometer are effected by magnets, are slow to respond and measure magnetic north (not true north). Infrared are cost-effective, but short range and saturated by sunlight. Ultrasonic range sensors are cost-effective and good for detecting large objects, but can't detect glass/water, only measure a few metres, have a wide beam and provide medium accuracy. RADAR uses radio (instead of sound) to detect objects at long distance, but are relative more expensive then ultrasonic range sensors. Image sensors (video cameras) are a cost-effective, rich in information, and two or more can get depth information, but are computationally expensive, hard to process the data (aka computer vision), affected by dust/fog/rain, and light variations. LIDAR are high accuracy (about 1mm), however are expensive (prices are coming down every year), but can't detect glass/water. Distance measurement sensors are easy to interpret, other sensors are hard. Colour constancy and object classification is very hard (e.g. “Is it a tree or a human?”).

We have a deep understanding of control systems. We typically use Linux-based SBC's and a program a custom PID controller - perhaps even a cascade PID controller, bayesian filters, particle filters, kalman filters, Monte Carlo methods, or train a deep neutral network. The outputs of these systems might control various types of motors (e.g. brushed, brushless, servo, steppers) and/or various types of actuators (e.g. linear, pneumatic, hydraulic), and/or other things like lights or speakers.

We have been involved with computer vision and machine vision since early 2000's - we where involved in the development of two computer vision libraries before OpenCV became popular. Have done much biologically inspired techniques such as optical flow. Was involved in the early days of artificial intelligence using techniques such as Local Binary Patterns (LBP) and Haar-like features (HAAR). These days we typically use machine learning methods such as designing and training deep neural networks (outstanding for vision-based object recognition using ImageNet).

We where involved in the development of a robotic operating system which had 364,578 lines of code, before ROS was written.

We have developed custom software for various manipulators, and have a good understanding of forward and inverse kinematics.

We appreciate that challenges with robotics - particularly with robot navigation, computer/machine vision, and manipulation with the real-world, in real-time using real-robots.

We are confident with a broad range of skills and confident our Melbourne Robotics Prototyping services can offer solutions such as:

• Autonomous Mobile Robots
• Simultaneous Localisation and Mapping (SLAM)
• Path Planning, Goal Recognition
• Collision Avoidance
• Pattern Recognition
• Object Recognition
• Object Tracking
• Genetic Algorithms
• Neural Networks
• Evolutionary Robotics
• Probabilistic Robotics
• Machine Learning
• Deep Learning
• Optical Flow, Point Clouds
• Stereo Active Vision, Panoramic Cameras, Fisheye Lens/Camera, Log-polar Camera, Time of Flight Cameras, Camera Array
• 3D Reconstruction
• Manipulator, Biped and/or Quadped locomotion
• Sensor Fusion: GPS, LIDAR, IR, Ultrasonic, Vision
• PID Controllers, Cascade Control Loops, Fuzzy Logic
• Bayesian Filters, Particle Filters, Kalman Filters, Monte Carlo Methods

These technologies can be used for various applications such as:

• Lightweight Reconnaissance Vehicle (LRV)
• Remotely operated Underwater Vehicle (ROV)
• Autonomous Underwater Vehicle (AUV)
• Unmanned Aerial Vehicle (UAV)
• Storm Drain and/or Pipe Inspection Robot
• Traffic Control Robot
• Gardening Robot
• Robot Lawn Mower
• Floor Cleaning Robot
• Home Land Security Robots
• Forklift Automation
• Load Haul Dump Automation
• Agriculture Automation
• Production Line Automation
• Asset Tracking
• Kid Tracking
• Pedestrian Tracking

Melbourne is the capital and largest city of Victoria, and the second most populous city in Australia. In 2008, it had a population of approximately 3.9 million. Melbourne is a centre for arts, commerce, education, industry, sports and tourism. Since 2002, it has been consistently ranked in the top three 'World's Most Livable Cities'.

Melbourne was named after the 2nd Viscount Melbourne, William Lamb in 1837; the Prime Minister of the United Kingdom during the reigns of both King William IV and Queen Victoria. It is located on the lower reaches of the Yarra River and on the northern and eastern shorelines of Port Phillip, extending into their hinterland.

It was established in 1835 (47 years after the first European settlement of Australia), by free settlers from Van Diemen’s Land, as a pastoral township around the estuary of the Yarra River. Melbourne was declared a city by Queen Victoria in 1847, and became the capital of Victoria when the district was declared a separate colony from New South Wales in 1851. When gold was discovered in the district during the 1850s (which sparked the Victorian gold rush) Melbourne was transformed into a wealthy metropolis, and one of the largest and richest cities in the world, by the 1880s.Upon the Federation of Australia in 1901, Melbourne served as the seat of government of the newly founded Commonwealth of Australia till 1927 while the new nation's capital of Canberra was being built.

The city is notable for its distinct blend of Victorian and contemporary architecture, expansive parks and gardens and multicultural society. It is also home to the World’s largest tram network. It is recognised as Australia's 'cultural and sporting capital'. In 2007, it was also ranked in the top five university cities in the Global University Cities Index by RMIT, and was classified as a City of Literature by UNESCO in 2008.