Glydways builds its first pedestrian detection model with Uber AI Solutions
Learn how Glydways, an autonomous vehicle company, established a scalable light detection and ranging (LiDAR) data pipeline to develop a critical perception capability.
2 September 2026 | United States
Executive summary
When autonomous mobility company Glydways needed to build its first pedestrian detection model, it partnered with Uber AI Solutions to establish a scalable LiDAR data pipeline that enabled the company to develop a critical perception capability and deploy in less than a quarter.
Key results
99%
Annotation accuracy achieved within two months
20,000+
LiDAR frames labelled
40%
estimated cost savings in labelling
Deployment requires reliable pedestrian detection
Glydways builds a new category of urban mobility called Flow Networks. It integrates purpose-built autonomous vehicles, real-time orchestration, and dedicated guideways into networks that move riders on-demand without stopping, delivering personal rides at transit fares.
“
We cannot deploy our system without reliably detecting pedestrians throughout our entire system.
”
Farshid Moussavi
Head of Perception, Glydways
With limited early-stage training data, every annotation mattered. Glydways needed more than labelling capacity. It needed an operational process capable of producing consistently high-quality training data as its perception models evolved.
As Moussavi explained, "3D labelling of LiDAR point clouds is complex, especially at longer ranges where fewer points exist on target."
Developing annotation standards, QA workflows and collaborative reviews
Rather than building that capability from scratch, Glydways partnered with Uber AI Solutions to create a scalable data operation.
“
Uber AI Solutions' quality, responsiveness, capabilities, internal tooling, and prior experience with 3D labeling at scale were all compelling.
”
Farshid Moussavi
Head of Perception, Glydways
Together, the teams developed annotation standards, quality assurance workflows and a collaborative review process that continuously refined labelling quality while improving efficiency. Workflow experiments helped identify the best balance between speed and precision, while recommendations around data collection and annotation strategy strengthened the overall training dataset.
“
Open communication and teamwork across teams were key to identifying issues early on and maintaining steady progress. Working with Uber AI Solutions is the smoothest experience I've ever had working with a data labeling partner.
”
Farshid Moussavi
Head of Perception, Glydways
Establishing a repeatable capability for perception development
The partnership produced more than a high-quality dataset. It established a repeatable operational capability for future perception development.
“
Uber AI Solutions enabled us to get our first externally labeled dataset for pedestrians. This led to our first internally trained and deployed pedestrian detector.
”
Farshid Moussavi
Head of Perception, Glydways
The engagement also created a structured pipeline for data ingestion, annotation and quality review that Glydways can continue to use as it expands into new perception scenarios.
Expanding into additional object classes and tasks
With a proven data pipeline in place, Glydways can confidently expand its perception capabilities into additional object classes and tasks, and increasingly complex operating conditions. Summing up the collaboration, Moussavi explained, "The most valuable result was the establishment of a reliable and quality labelling pipeline and partnership we can count on for future tasks."
“
Uber AI Solutions is a force multiplier and extension of our team that is critical to delivering a safe, reliable ML-based system.
”
Farshid Moussavi
Head of Perception, Glydways
For Glydways, the engagement wasn't simply about labelling data. It established the operational foundation needed to build, improve and scale perception models as the company's autonomous transit platform continues to evolve.
Learn how Uber AI Solutions can partner with your enterprise to further your AI.
Testimonials reflect individual results. Individual results do not guarantee the outcome for any given customer, and customer experience will vary.