First look: Terrain Intelligence 2024 dashboard
12 December 2024
This is the first time our Terrain Intelligence dashboard analysis tool has been seen outside our company and project partners. Explore how our AI platform is delivering customer-led features off-road and on-road.
We recently announced major updates to Terrain Intelligence, our AI-driven software platform that uses a single camera and tiny compute capacity to read surface characteristics in unstructured, unmapped environments.
Here, for the very first time, we’re pulling back the curtain on our 2024 dashboard analysis tool for Terrain Intelligence. This shows how our modular AI models can deliver customer-led features across different sectors and vehicle types.
Key features walkthrough
We use this Terrain Intelligence dashboard with our project partners to visually represent the outputs our modular AI technology is delivering. The main features include:
Camera view. Terrain Intelligence reads from a single, conventional automotive mono-camera facing the road ahead. In this use case, the camera is fitted to a motorcycle. Terrain Intelligence is reading surface characteristics in advance of the bike and creating “flags” that can act as triggers for vehicle systems based on tuneable parameters.
Switchable prediction parameters. In this configuration, we’ve set a minimum prediction time threshold, rather than distance. Why is this useful? Different vehicle settings (such as opening and locking diff locks or altering damper firmness) have different latencies. Providing sufficient time to alter a vehicle setting is the most important data in this use case, and that’s what this setting achieves.
Triggers. In this use case, we have set prediction flags (denoted by coloured dots on the charts) to read for upcoming “suspension events.” These are surface features that would negatively affect rider comfort or vehicle stability, such as divots, cracks, potholes, bumps and boulders. The flags in this case can trigger adjustments to semi-active suspension to achieve pre-determined outcomes in comfort, stability or performance. However, flags can be set for a wide range of surface characteristics and vehicle systems, not just suspension.
Verification. In the charts, you can see our prediction flags overlaid onto vehicle IMU data to assess real-world accuracy. Everything to the right of the vertical blue lines is predictive (ie ahead of the vehicle); the blue line on each chart represents the vehicle and its IMU reaching that location and beginning to generate independent data. We can and do integrate additional sensors other than IMU during validation work, depending on the needs of our customers.
A new generation of customer-led features
Terrain Intelligence unlocks a new horizon in off-road and on-road ADAS systems that dramatically reduce cognitive load for drivers. Here are some predictive features that our software can enable:
Driver alerts, including warnings about hazards and suggestions for drive modes
Proactive suspension, such as changing damper settings or ride height in advance of changes in surface type, gradients or hazards, or even moving wheels to maintain or avoid surface contact in advance of terrain features
Hill recognition, including incline start, crest, and decline start and end, married to drive mode and chassis adjustments, such as engaging four wheel drive, engaging low range gears, locking differentials etc
Vehicle to vehicle communication - our flags could be combined with the vehicle’s existing GPS and data links to inform other vehicles of impending hazards (Terrain Intelligence does not require live data links for operation)
Speed selection, helping a vehicle slow down or speed up to match terrain changes ahead
Near or far: which is better?
While the use case in this walkthrough is set for prediction in time, we can also select for distance. We help our customers strike the right balance here for their needs.
Reading the surface closer to the vehicle - what the vehicle is very likely to hit - allows us to minimise compute load and maximise accuracy. We believe this is the “sweet spot”, as it’s also the range where a driver might see an obstacle but not be able to react in time.
However, if a customer wants to provide route planning to reassure drivers that the environment ahead is passable, we can tune Terrain Intelligence to look further ahead. Because this requires more compute-intensive activity, it is more suitable in purpose-built software-defined vehicles with a lot of spare computational capacity, such as the latest fully-autonomous vehicles.
Rigorous testing to refine performance
We’ve learned a lot from deploying Terrain Intelligence on different test vehicles over the past 12 months, including motorcycles, side-by-sides, pick-up trucks, SUVs and even an armored truck, not to mention our own all-electric, quad-motor platform, P0.
We’ve carried out testing in a wide range of different off-road environments, including dirt, gravel, mud and sand. We’ve been developing Terrain Intelligence capabilities on the road too, to identity features such as potholes, speed bumps, drain covers, expansion joints and rough asphalt, and at high speeds (up to 100km/h). The result is a high level of accuracy and wide operating windows.
See our tech stack live at CES 2025
Learn more about Terrain Intelligence at CES (6-10 Jan 2025, Las Vegas). We’ll present our integrated technology on our motorcycle demo vehicle at the Connected Vehicle Systems Alliance (COVESA) evening event at The Bellagio on 7 January 2025.