Research for AI that works where people live.
Most AI is trained for clean data and ideal conditions. Our research starts from the messy real world: unmarked roads, mixed traffic, limited hardware and no guaranteed connection.
Self-Supervised Speed Breaker & Pothole Anticipation
Learning to see unmarked road hazards before the vehicle feels them.
- The problem
- Speed breakers and potholes on local roads are often unmarked and hard to see, especially at night. Drivers find them only when the vehicle hits them.
- Our approach
- A camera model that learns from the vehicle's own motion signals, so it can be trained on everyday driving without hand-labelling thousands of images.
- Where it's going
- An on-device warning system that runs in real time on low-cost hardware.
Motorcycle-Aware Collision Warning
Tracking motorcycles and rickshaws to warn before a collision.
- The problem
- Most collision-warning systems are designed around cars. In South Asian traffic, motorcycles and rickshaws are the majority, and they move unpredictably.
- Our approach
- Detection and multi-object tracking tuned for small, fast two- and three-wheelers, predicting their paths to estimate collision risk.
- Where it's going
- A driver-assistance module that warns early enough to react.
Four directions, one goal.
Embedded AI
Compact perception models that run on-device, in real time and offline, on hardware that is affordable to deploy at scale.
Robotics
Sensing, control and decision-making for machines that move through and act on the physical world.
General AI
Applied machine learning, computer vision and language models, from experiments to published work.
AI Automation
Agents and intelligent workflows that take on real, repetitive business processes reliably.
Our principles
Real conditions first
We collect and test on the roads, traffic and environments our systems will actually face.
On-device by default
If it needs a data centre to work, it won't help most people. We design for the edge.
To the end solution
A paper is a milestone, not the finish line. Every project aims at something people can use.
Publications
- ForthcomingSelf-Supervised Speed Breaker & Pothole AnticipationEmbedded AI
- ForthcomingMotorcycle-Aware Collision WarningEmbedded AI
Research with us.
We welcome partnerships with universities, industry and public bodies working on road safety, robotics and edge AI.