Spatial Reasoning Evaluation Specialist - Video Reinforcement Learning

Role Overview

We are seeking experts to join as Spatial Reasoning Evaluation Specialists within our Video Reinforcement Learning (Video RL) initiative. This position requires assessing AI-generated video outputs, inspecting autonomous agent trajectories, and validating physical and temporal plausibility along with spatial reasoning aspects.

Primary Duties

• Perform detailed frame-level analysis of autonomous video agent behaviors to identify inaccuracies related to physics, causality, and execution anomalies.
• Verify multi-event sequences to confirm correct chronological order and adherence to temporal logic.
• Assess whether the behaviors of autonomous agents correctly pursue predefined multi-step goals, pinpointing any deviations or inadequate step execution.
• Validate spatial elements including bounding boxes, optical flow fields, pixel-wise semantic segmentation, and skeletal animations of avatars.
• Develop comprehensive and structured failure mode reports categorizing issues such as visual anomalies, unrealistic physics representations, and graphical clipping errors.
• Create robust ground-truth datasets and compile edge-case scenarios to enhance the precision of automated verification methods.

Candidate Qualifications

• Possession of a Bachelor's or Master's degree in fields like Computer Science, Engineering, Physics, Mathematics, Data Science, or other STEM disciplines or equivalent practical experience.
• Familiarity with video-centric analysis methods, computer vision, reinforcement learning, robotics, or simulation platforms.
• Exceptional spatial and temporal reasoning capabilities.
• High attention to detail with capability to conduct intensive evaluation activities.
• Proficiency in English, both written and verbal, with strong reasoning skills.
• Basic Python programming knowledge is advantageous.
• Experience or knowledge of AI/ML concepts and proficiency with advanced video annotation software such as CVAT, Labelbox, or Roboflow is preferred.