Yondu is creating the robotic workforce of the future starting with logistics automation. We're deploying humanoid robots in the first flexible, drop-in picking automation solution designed for 3PLs. Our approach is building robot foundation models to enable off the shelf robots to do these tasks autonomously. To train these models, we're building our own teleoperation system in house. Finally, there's a whole stack of classical robotics and integration work to make our systems work in the real world . Learning constantly and keeping up with SOTA research is one of the pillars of our company's culture. The company was founded by MIT students and has raised seed funding after graduating from the YC W24 batch. Join us in shaping the way the world works.

Note: Above salary is assuming 3 months

About Us:
Yondu is deploying wheeled humanoid robots to automate logistics and warehouse operations. Our teleoperation system enables human operators to remotely control robots in real time while collecting the data needed to train autonomous behaviors.

We’re now expanding this system with advanced real-time 3D rendering and VR to make remote operation more intuitive, especially for objects and environments that aren’t pre-mapped or pre-modeled.

Role Overview:
We are looking for a Computer Vision Intern to lead a focused project on exploring methods for real-time 3D scene reconstruction and VR visualization for our teleoperation platform.

You’ll be responsible for designing and implementing systems that model the “unmodeled” — anything not already represented in our static robot, shelf, bin, or product models — to give operators richer, more accurate situational awareness. This role blends computer vision, graphics, robotics, and VR development in a fast-paced R&D environment.

Key Responsibilities:

• Delve into state of the art research
• Develop real-time 3D reconstruction pipelines from sensor data (RGB-D, stereo, LiDAR) for teleoperation environments.
• Integrate reconstructed scenes into VR/AR interfaces to improve operator perception.
• Implement dynamic modeling of unknown or moving objects in the robot’s workspace.
• Work closely with our robotics and perception teams to interface with ROS-based systems.
• Optimize rendering performance for low-latency teleoperation.
• Explore and prototype new VR-based UI/UX concepts for remote control.
• Deploy and test your work on real robot hardware (Jetson platforms, ROS2).

Requirements:

We are looking for applicants who have done graduate level research or have industry experience implementing similar projects.

• Strong software background
  • Proficiency in Python, C++, and/or C#
  • Understanding of networking protocols for real-time systems
  • Experience setting up and maintaining networked servers (for VR or teleop)
• Unity experience
  • Building VR applications and interactive 3D environments
  • Integrating real-time sensor/vision data into Unity scenes
• Computer vision / 3D data processing
  • OpenCV, Open3D, or similar libraries
  • Real-time point cloud, mesh, and texture processing
• Computer graphics
  • Rendering pipelines, GPU acceleration, CUDA optimization
• Robotics experience
  • Hands-on work with ROS (preferably ROS2)
  • Experience with Jetson platforms or other embedded systems
  • Familiarity with multi-sensor calibration and integration

Bonus Skills:

• Experience with SLAM, visual odometry, or real-time reconstruction algorithms
• Knowledge of HMDs (Meta Quest, HTC Vive) and VR SDKs

What You’ll Get:

• Direct experience working on real humanoid robot teleoperation systems
• Ownership of a high-impact project from concept to deployment
• Opportunity to push the boundaries of robotics + VR integration
• Mentorship from a stacked team building cutting-edge embodied AI systems

Timeline and Compensation:

The compensation will be $3k-$5k a month. The salary below is monthly. The standard length will be 3 months with an option to extend it to a full Co-Op.

Our approach is building robot foundation models to enable off the shelf robots to do these tasks autonomously. To train these models, we're building our own teleoperation system in house. Finally, there's a whole stack of classical robotics and integration work to make our systems work in the real world . Learning constantly and keeping up with SOTA research is one of the pillars of our company's culture.