Senior Perception Learning Engineer

Description

The company builds AI-powered humanoid robots meant to work alongside people. Its first target industries are manufacturing and logistics, with healthcare and home use planned later.

The team is preparing the company's humanoid for commercial launch and volume production, which means working through safety, commercialization and manufacturing scale. The work spans the whole robotics software stack.

In this role you will advance perception systems that let the humanoids understand and act in busy human spaces. The focus is SLAM, visual-inertial odometry, world modeling, learning-based perception, object detection and multi-sensor fusion, which together form the base for dependable autonomy outside the lab.

You will build and tune deep learning models for real-time detection, tracking, segmentation, scene understanding and state estimation, and help build the pipelines used to train, evaluate and deploy them. You will also merge camera, LiDAR, depth and IMU data into unified world models used for navigation, manipulation, safety and human-robot interaction.

The role balances research with engineering, so the results must run well in production. You will work with the reinforcement learning, platform software and systems teams and help set the direction of the perception and autonomy roadmap.

Responsibilities

• Lead the design, development and tuning of perception and SLAM pipelines for humanoids, covering visual-inertial odometry, mapping, localization, detection, tracking, segmentation, pose estimation and scene understanding.
• Build multi-sensor fusion frameworks that combine cameras, LiDAR, depth sensors and IMUs into reliable real-time state estimation and mapping in dynamic settings with people.
• Help create scalable data pipelines, training infrastructure and inference frameworks that speed up model development, evaluation and deployment.
• Research and ship learning-based models for SLAM, 3D scene understanding and perception tuned for locomotion, manipulation and human-robot interaction.
• Set up performance profiling, regression tests and telemetry so perception and SLAM modules meet strict latency, accuracy and reliability targets on edge devices.
• Work with planning, control and hardware teams to define perception-to-action interfaces that run in real time with locomotion and manipulation.
• Guide the use of synthetic data, such as from simulators like IsaacSim, alongside real-world datasets to improve model generalization and robustness.
• Help set standards for code quality, model versioning, reproducibility and deployment.

Requirements

• MS or PhD in computer science, robotics, computer engineering or a related field.
• 3-5+ years of experience developing and deploying perception or SLAM systems in robotics or real-time vision.
• Solid grounding in SLAM, visual-inertial odometry, probabilistic state estimation and deep learning for computer vision.
• Practical skill in detection, segmentation, multi-object tracking, 3D perception and learning-based mapping and localization.
• Hands-on use of PyTorch, JAX or TensorFlow, plus vision and multi-modal tools such as OpenCV, Detectron2, YOLO and foundation models like SAM, CLIP, DINOv2 or Flamingo.
• Proficient in Python and modern C++, with strong engineering habits in version control, testing and CI/CD.
• Working knowledge of 3D geometry, camera models and estimation methods such as EKF, UKF and factor graphs.
• Experience deploying optimized perception or SLAM models on edge hardware, including GPUs, NPUs and embedded platforms, under tight compute, latency and thermal limits.
• A record of taking ML, perception or SLAM work from research into production robots.
• Preferred: experience with humanoid robots, bipedal locomotion and manipulation.
• Preferred: experience with SLAM frameworks such as ORB-SLAM, VINS, Kimera or Cartographer, or with learning-based SLAM.
• Preferred: strong classical computer vision, including geometry-based methods and feature extraction.
• Preferred: skill in model acceleration, quantization or compression using TensorRT or ONNX Runtime.
• Preferred: familiarity with ROS 2, GStreamer or zero-copy pipelines.
• Preferred: knowledge of synthetic data generation and domain adaptation for perception and SLAM training.
• Preferred: contributions to open-source robotics or vision software.

Conditions

Annual salary range is $220,000 to $250,000.

Direct hire only. Agencies should not submit candidates.