Research areas
In addition to providing engineering, solutions, and development services to our clients, we also conduct our own internal research. We strive to stay on the cutting edge of technology and develop custom solutions and systems. This page summarizes some of the general research areas we participate in. We always welcome collaborations and partnerships in these and other areas!

We've also provided access to some of the papers published by our team members, where the research has been sponsored by Aptus.

AI Medical Diagnostics
Aptus is a partner in Multus Medical - a medical technology company that develops 3D renderings of patient-specific anatomy based on MRI scans. We developed all operational technology and software for Multus.

A portion of the work we did for Multus included the development of Artificial Intelligence (AI) radiology software. This software takes in Magnetic Resonance Imaging (MRI) scans of patient's spine, and generates a radiology report that indicates abnormalities and pathologies it identified in the spine. These include central canal stenosis, foraminal stenosis, abnormalities to the lordotic curvature, etc.

The software takes in slices of MRI images in series in DICOM format, performs 3D segmentation of the patient's spine to identify anatomy - discs, spinal cord, etc., crops regions of interest for each disc and runs them through AI detectors to identify pathologies and abnormalities. It then generates a radiology report comparable to those generated by radiologists. Below is a very high level diagram showing this.
In partnership with Multus and the Center for Advanced Spine Care of Southern Arizona in Tucson, we conducted feasibility and reliability studies on the AI we developed against surgical outcomes and human board-certified radiologist reads.

We were able to show that the AI radiology bot performed on par and sometimes even better than radiologists. This is a revolutionary tool for the medical industry, and we are working closely with Multus to make sure as many get the benefit of this as soon as possible!

We have submitted (currently in editorial process) a few of these studies to the International Journal of Spine Surgery. Below are the manuscripts.

Feasibility of Deep Learning Algorithms for Reporting in Routine Spine Magnetic Resonance Imaging
Submitted to International Journal of Spine Surgery
[2.6 MB]
Reliability Analysis of Deep Learning Algorithms for Reporting of Routing Lumbar MRI Scans
Submitted to International Journal of Spine Surgery
[364.9 kB]
Probability of Predicting Successful Transforaminal Endoscopic Decompression for Herniated Disc and Foraminal Stenosis with Deep Learning Algorithms Interpreting Lumbar MRI Scans
Submitted to International Journal of Spine Surgery
[734.5 kB]

Space Robotics
A lot of sponsored research has been conducted in the field of space robotics by Narendran Muraleedharan (Naru) - a founder and core developer at Aptus. The research is focused on using model-free or model-minimalistic methods to emulate a free-floating environment using a robotic platform. Such an emulated environment can be used for extensive testing of space robots on Earth prior to launch. The system allows for testing of orbital control systems, attitude systems, camera systems, docking procedures, and other sub-systems. Testing these systems on Earth saves companies and space agencies billions of dollars in failed launches.

Complex dynamic coupling between the satellite bodies and robotic manipulators on space robots make it very difficult to accurately emulate movement of the robot. Generally, Hardware in the Loop (HITL) simulation is used, however accurate knowledge of the system model and dynamic parameters is necessary for such a simulation. This research proposes a gravity compensated force-feedback control method and a center of mass regulator that allows for the emulation of a free-floating environment with minimal (for spatial) or no (for planar) knowledge of the system model or dynamic parameters.

Following are a few of the publications that resulted from this research venture.
Recreating Planar Free-Floating Environment via Model-free Force-Feedback Control
Presented at the IEEE 2016 Aerospace Conference
[1.6 MB]
Development of an Emulated Free-Floating Environment for On-Earth Testing of Space Robots
Master of Science Thesis by Narendran Muraleedharan
[22.8 MB]
Experimental Validation of a Planar Free-Floating Emulator via Model-free Force-Feedback Control
Presented at the IEEE 2018 ICMA Conference in China
[520.2 kB]

Spherical Robots
Generally, mobile robots use differential drive for locomotion. However, differential drive robots are not favorable for exploration in rough terrain due to the possibility for wheels to get stuck in harsh environments. Spherical robots have been tested in the past for exploration and locomotion in uneven terrain.

Spherical robots have used various different methods of locomotion. These include a weighted pendulum, internal drive (like a hamster ball), control moment gyroscopes, and momentum wheels. One of the simplest methods for locomotion is using a weighted pendulum. In the past, companies have used a drive-and-steer system with the internal pendulum. However, such a system does not allow for omnidirectional locomotion - which is one of the most attractive features of a spherical robot.

This research focuses on developing a control system that allows for omnidirectional movement. Such a controller was developed and the results were presented at the IEEE 2016 Southeast Conference in Norfolk, Virginia.

Below is a publication on the designed omnidirectional controller for spherical robots.
Omnidirectional Locomotion Control of a Pendulum Driven Spherical Robot
Presented at the IEEE 2016 Southeast Conference
[955.5 kB]

Traffic Flow Optimization
A recent project undertaken by the Aptus team in collaboration with Bjorn Forsdal, is aimed at improving traffic flow with the use of smart intersections and artificial intelligence. Many intersections are already equipped with traffic sensors including cameras, trip sensors, and weight sensors under the road. This research explores various options for the use of artificial intelligence in order to optimize the traffic flow pattern and minimize commute times.

We have developed a traffic simulator in order to test custom systems. We're training a Long Short-Term Memory (LSTM) neural network with auto-generated training data and brute force optimization in order to produce more effective logic for traffic light sequences. We are also experimenting with genetic and evolutionary algorithms to train the neural network.

This is an ongoing research project. More information on the project will be posted as it becomes available.