top of page
  • Sam Claassens

Next Generation Robotics

We’re all excited for the Jetson-like world with robotics in every aspect of life, solving a variety of challenging problems. But most are failing to leave safe, controlled lab spaces. Why?

The Imperfect Information Problem: Operating in the real-world requires robust solutions that can readily manage the chaos of dynamic environments and imperfect data. In the case of robotics, this is the deciding factor between a commercially-successful robot and a failure to launch.

Why Marine Vehicles: Coordinating marine vehicles is an ideal example of automation in a complex, dynamic environment. How do you ensure that your vehicles can safely track and operate around other ships, make the most of the variety of potentially-disagreeing sensors, and robustly handle the busy environment while completing critical tasks?

The NavAbility Case Study: At NavAbility we’re using data from MIT SeaGrant‘s  REx/Philos vehicles to demonstrate how any robot can extract map information from multiple sensors, identify and track dynamic objects like ships, and use prior information to navigate effectively in a dynamic environment.

Robotics subject to GPS-denied operations

Marine surface vehicles benefit from GPS information. However true automation requires complete situational awareness by integrating all the sensor data – such as radar and camera information – into a comprehensive map of the current environment. 

At NavAbility, we flipped the problem around:

  • Can we produce a complete map using just radar data? Why? This is the “hard” problem and we wanted to demonstrate this first. It’s also relevant to subsurface vehicle applications as well as high-resolution surface vehicles positioning. GPS can easily be added if available, improving the solution’s accuracy.

  • Can we utilize multiple data sources (such as camera feeds and radar)? A robust object identification and tracking system will make use of the map and all accompanying sensor data for the best solution.

  • Can we share this map between vehicles for coordinated swarming? A shared map is significantly more useful than isolated maps on each vehicle.

Imagine if the various sensors are combined into a single source of truth that:

Robustly ignores imperfect data by considering all hypotheses (important with radar data)

  1. Includes both the vehicle’s location as well as position+velocity estimates of all nearby obstacles

  2. Streams the data into a shared map that can be consumed by all vehicles in the environment

  3. Continuously improves as more data is incorporated into the map

  4. Publishes updated estimates of all objects in the environment

  5. Provides complete situational awareness for all path planning systems

See Full Case Study Here


bottom of page