Bedrock Robotics
Machine Learning Engineer: Perception Analytics
About this role
Bedrock is seeking a Machine Learning Engineer to develop perception algorithms that extract actionable analytics from construction site sensor data. You'll build real-world computer vision systems for vehicle re-identification, productivity metrics, and safety monitoring across autonomous construction equipment deployments.
What you'll do
- Design and implement perception algorithms for multi-camera vehicle re-identification on large construction sites
- Adapt and extend existing perception models to generate customer-facing analytics metrics
- Collaborate with customers to define feasible perception requirements and analytics capabilities
- Deploy trained models and analytics pipelines to production fleets of autonomous machinery
- Perform data analysis to characterize sensor behavior, identify edge cases, and design evaluation metrics
- Integrate raw sensor inputs (camera, lidar, IMU) into end-to-end learned pipelines
What they're looking for
- Computer vision and object re-identification (re-ID)
- Deep learning frameworks (PyTorch preferred)
- Python programming
- Systems languages (C++ or Rust)
- 3D geometry, sensor calibration, and coordinate transforms
- Raw sensor data processing (camera, lidar, IMU)
- Statistical analysis and anomaly detection
- Production deployment of perception systems
Opens the official application on the employer’s site. No login required.
Bedrock Robotics
Bedrock Robotics develops autonomous construction machinery powered by AI and robotics technology. The company is hiring for roles spanning developer infrastructure, simulation systems, hardware engineering, field robotics application, and frontend engineering to support the development and deployment of autonomous excavators and heavy equipment.
View all jobs at Bedrock RoboticsLikely interview questions
- Walk us through a perception system you've shipped to production—what were the biggest challenges moving from lab to deployment?
- How would you approach re-identifying the same vehicle across multiple camera views when lighting and environmental conditions change significantly?