Robotics Research Engineer

Abaka AI•Mountain View, CA

61d•$100,000 - $140,000

About The Position

Abaka AI is hiring a Robotics Research Engineer for Abaka Robotics Lab, its in-house research group focused on building the data foundation for physical AI. This hands-on, research-driven role combines robot learning, data production, and publishable research. The engineer will train and evaluate policies in simulation and on real-world robotic systems, use experimental results to understand which data matters, and design scalable pipelines that turn raw sensor data into high-quality training datasets with minimal human intervention. The role involves working with 6-DoF collaborative robot arms and an ALOHA-style bimanual platform with full teleoperation. Findings will inform datasets built for Abaka AI’s partners and contribute to publications at leading robotics, computer vision, and machine learning venues.

Requirements

MS or PhD in Robotics, Computer Science, Machine Learning, or a related field.
1–3 years of hands-on experience with real robot hardware through research, an internship, or a project.
Strong machine learning fundamentals, including optimization, generalization, evaluation methodology, and end-to-end model training.
Strong robotics fundamentals, including coordinate frames, kinematics, and basic control.
Strong 3D vision fundamentals, including camera models, multi-view geometry, and point clouds.
Rigor in evaluation; claims about a policy or dataset must be backed by measurements.

Nice To Haves

Publications or workshop papers at top-tier robotics venues (CoRL, RSS, ICRA, IROS), vision venues (CVPR, ICCV, ECCV, 3DV), or ML venues (NeurIPS, ICML, ICLR); a widely used open-source repository or a dataset adopted by other teams is valued equally.
Experience with imitation learning or VLA codebases such as LeRobot, ACT, Diffusion Policy, OpenVLA, or π0.
Experience deploying a learned policy on a real robot and conducting rigorous evaluation.
Practical 3D reconstruction experience with COLMAP, SLAM, 3DGS/NeRF, or point-cloud registration.
Experience with world models or video generative models, particularly as simulators or data generators.
Experience with RL fine-tuning of learned policies or model-based RL.
Experience with hand and object pose estimation, hand-object interaction, or human-to-robot motion retargeting.
Depth in a simulator such as MuJoCo, Isaac Sim/Lab, Genesis, or SAPIEN, including environment and asset authoring.
Experience with domain randomization, procedural scene generation, or sim-to-real transfer.
Experience with teleoperation or data collection systems in the ALOHA, UMI, or hand-tracking lineage.
Familiarity with egocentric datasets such as Ego-Exo4D, EgoMimic, HOT3D, Project Aria, and DexCap.
Experience in dexterous or bimanual manipulation research.
Experience with camera calibration and multi-sensor time synchronization.
Experience applying vision-language models to robotics, including task planning, language-conditioned policies, and reward or success detection.
Experience with auto-labeling or ground-truth generation over real sensor data, with quality metrics.
Experience with tactile, force-torque, or other multimodal sensing for manipulation.
Experience with data curation and dataset-scaling studies, including deduplication, mixture ratios, difficulty or diversity scoring, and ablations that identify which data mattered.

Responsibilities

Train and evaluate robot policies—including imitation learning, diffusion policies, VLA, and RL fine-tuning—using Abaka AI data and public baselines in simulation and on real robots.
Use evaluation results to assess data quality and determine which data helped policy performance, which did not, and why.
Reproduce relevant research papers, run current methods on the company’s hardware, and develop findings suitable for publication at leading robotics, vision, or machine learning venues.
Maintain training and evaluation infrastructure, including training runs, evaluation harnesses, and simulation or world-model environments used before hardware tests.
Design pipelines that convert raw captures—such as egocentric human video, robot logs, teleoperation, and 3D scans—into training data with minimal human labeling.
Synthesize missing data, automate labeling with models and geometry, and curate data for training.
Track how data decisions affect policy performance.