Role Overview

VinRobotics is seeking Senior Computer Vision Engineers to join our Humanoid Robotics Perception Team. You will design and deploy real-time perception systems enabling humanoid robots to see, localize, understand, and navigate complex real-world environments.

This role focuses on 4 core perception pillars:

1. Object Pose Estimation & Grasp Perception for manipulation

2. Stereo Depth Estimation for geometry understanding

3. Semantic Segmentation & Mapping for navigation and scene understanding

4. Visual SLAM

Your work will directly power grasping, manipulation, obstacle avoidance, semantic mapping, visual SLAM, and autonomous navigation on next-generation robotic platforms.

Key Responsibilities

6-DoF Object Pose Estimation & Dexterous Grasping (Manipulation Perception)

• Design and implement 6D object pose estimation pipelines using RGB, RGB-D, or stereo inputs

• Handle occlusion, symmetry, cluttered scenes, and domain shift

• Integrate pose outputs with grasp planning and manipulation stacks

• Optimize inference pipelines for real-time robotic execution

• Develop grasp detection and 6-DoF grasp pose synthesis models for multi-finger (dexterous) robotic hands.

• Estimate object shape, affordances, and contact points to support in-hand manipulation and fine motor control.

• Integrate perception outputs with grasp planning, force/tactile feedback, and multi-finger control stacks.

• Handle challenging objects, including deformable, transparent, reflective, and small or thin items.

• Leverage sim-to-real domain adaptation and synthetic grasp datasets to improve robustness and generalization.

Stereo Depth Estimation

• Develop and optimize stereo depth estimation pipelines

• Handle challenging conditions:

  1. Low texture

  2. Reflective / transparent surfaces

  3. Outdoor / indoor lighting variation

• Evaluate depth accuracy, completeness, and latency under real robotic constraints

Semantic Segmentation & Navigation Perception

• Build semantic segmentation models for:

  1. Traversability

  2. Obstacle classification

  3. Scene understanding (floor, walls, objects, humans, dynamic agents)

• Contribute to semantic 3D mapping and semantic SLAM pipelines

• Support downstream modules such as:

  1. Local planning

  2. Obstacle avoidance

  3. Global navigation and relocalization

Sensor Fusion & System Integration

• Develop multi-camera perception systems (RGB, stereo, RGB-D)

• Integrate perception modules with ROS 2 Humble and real robot stacks

• Collaborate with SLAM, control, and motion planning teams (MoveIt, Nav2)

• Ensure robust synchronization, calibration, and frame alignment

Visual SLAM

• Design, implement, and optimize Visual / Visual-Inertial SLAM pipelines for real-time robot localization and mapping.

• Integrate loop closure and place recognition to ensure long-term localization consistency.

• Fuse multi-sensor data (RGB, stereo, RGB-D, IMU) for improved accuracy and robustness.

• Optimize SLAM systems for low latency, high reliability, and real-time deployment on robotic platforms.

• Integrate Visual SLAM with the navigation stack to enable fully autonomous localization, mapping, and path planning for humanoid robots.

Technical Requirements

Core Skills (Required)

• Strong background in Computer Vision, Robotics, or Deep Learning

• Solid understanding of multi-view geometry, epipolar geometry, and camera models

• Hands-on experience with deep learning frameworks:

  • PyTorch / Tensorflow

  • ONNX / TensorRT

  • CUDA (deployment & optimization)

• Experience with 3D data processing:

  • Open3D, PCL

  • NumPy, PyTorch3D

  • Proficiency in Python and/or C++ on Linux

Robotics & System Experience

• ROS 2

• Experience with camera drivers & sensors:

  • Intel RealSense

  • ZED (stereo & RGB-D)

• Familiarity with MoveIt / Nav2 / robotic execution pipelines

Preferred Qualifications

Bachelor’s or Master’s degree in Computer Vision, Robotics, AI, or related fields

Experience in one or more of the following:

• Multi-view perception

• Visual SLAM / Visual-Inertial systems

• Robot grasp learning

• Semantic mapping or navigation perception

• GPU optimization

• Distributed training

• Synthetic-to-real domain adaptation

What We Offer

Work on cutting-edge humanoid and autonomous robotics systems

Real-world deployment on state-of-the-art robotic hardware

Collaborative environment with AI researchers, roboticists, and system engineers

Access to GPU clusters, simulation environments, and large-scale datasets

Competitive compensation, benefits, and career growth opportunities