What are the responsibilities and job description for the Robot Learning Intern — Manipulation Policy Learning (2026 Fall) at Anyware Robotics position at Berkeley Industrial Engineering & Operations Research?
Job Title: Robot Learning Intern — Manipulation Policy Learning (2026 Fall)
About Anyware
Anyware Robotics builds general-purpose mobile manipulator robots for industrial applications. Our robots are deployed in real warehouse and logistics environments, supporting applications such as truck unloading, mobile palletizing, and machine tending.
The Role
The robots run on AnywareOS, our industrial physical intelligence system. We are looking for a Robot Learning Intern to work on applied manipulation learning using real robot data collected from production and in-house operations.
This role is designed for a strong, independent intern who can take a scoped robot learning problem from data understanding to model training, evaluation, and technical recommendation. Example directions include vision-language-action models, imitation learning, diffusion policies, action prediction, failure-mode analysis, or policy evaluation using multimodal robot data.
What You’ll Do
Work with real robot datasets, including various sensory inputs and system information related to robot behavior and outcomes.
Develop and evaluate robot learning models for a range of manipulation-related tasks.
Design, implement, and deliver approaches that bridge exploration and production readiness with imitation learning, diffusion policy, VLA models, or representation learning.
Build data pipelines, training scripts, evaluation metrics, and experiment reports.
Analyze successful and failed robot trials to identify learnable patterns and production-relevant failure modes.
Collaborate with planning, perception, and controls engineers to understand how learned models could augment the production system.
Present clear technical findings, including what worked, what failed, and what should be tested next.
Required Skills
Current MS or PhD student in robotics or machine learning.
Experience with robot learning, imitation learning, reinforcement learning, diffusion policies, VLA models, or visuomotor policy learning.
Strong Python programming skills and hands-on experience with PyTorch, JAX, or similar ML frameworks.
Strong data debugging skills, including dataset inspection, failure analysis, and metric design.
Familiarity with ROS, ROS2, robot logs, RGB-D data, point clouds, or robot kinematics.
Comfortable reading papers, implementing baselines, running experiments, and communicating technical tradeoffs.
Nice to Have
Experience working with real robot data, not only simulation.
Prior work on manipulation, grasping, or mobile manipulation robotics.
Experience with large pretrained models, vision-language models, or multimodal representation learning.
Publications, open-source projects, or strong course/research projects in robot learning.
What Success Looks Like
By the end of the internship, the intern should deliver a clear technical result: a trained model, reproducible experiment pipeline, evaluation report, and recommendation on whether the explored method is promising for Anyware’s production manipulation roadmap.
The goal is not just a demo. The goal is to deliver a validated model or method that demonstrates measurable improvement on defined manipulation tasks, along with clear evaluation results and a concrete recommendation for integration into the production system.
Why Anyware
An A Team: Anyware is built by top talents. Collectively, our team has won six Best Paper Awards and Finalists, including the ICRA 2025 Best Paper Award and ICRA 2024 Best Paper Award.
An A Robot: Our robot, Pixmo, won the Best Innovation Award (top-1) at both ProMat 2025 and MODEX 2026 — the two largest logistics automation exhibitions in the world. This is the first time in history that a robotics company has won this prestigious award back-to-back. Pixmo also won the RBR50 Robotics Innovation Award, recognizing it as one of the top 50 most innovative robotics solutions globally.
Real Impact: You will contribute directly to the core manipulation intelligence that powers every Pixmo deployment. Your work does not stop at a demo — it ships into real customer sites and solves real problems at scale.
Fast Growth: You will sharpen your skills with strong ownership and real responsibility. The team size is doubling every year, creating plenty of room for talents to step up and shine.
Compensation & Benefits
The expected salary range for this role is $30-60/hr, depending on experience and qualifications.
Benefits include daily meal, per diem, and unlimited snacks and beverages in the office.
Equal Opportunity Statement
Anyware is an equal opportunity employer. We are committed to creating an inclusive workplace and welcome applicants from all backgrounds. Employment decisions are based on qualifications, merit, and business needs.
How To Apply
Please submit your resume and any relevant materials through the Anyware Career platform. We look forward to learning more about your experience and interest in joining our team.
About Anyware
Anyware Robotics builds general-purpose mobile manipulator robots for industrial applications. Our robots are deployed in real warehouse and logistics environments, supporting applications such as truck unloading, mobile palletizing, and machine tending.
The Role
The robots run on AnywareOS, our industrial physical intelligence system. We are looking for a Robot Learning Intern to work on applied manipulation learning using real robot data collected from production and in-house operations.
This role is designed for a strong, independent intern who can take a scoped robot learning problem from data understanding to model training, evaluation, and technical recommendation. Example directions include vision-language-action models, imitation learning, diffusion policies, action prediction, failure-mode analysis, or policy evaluation using multimodal robot data.
What You’ll Do
Work with real robot datasets, including various sensory inputs and system information related to robot behavior and outcomes.
Develop and evaluate robot learning models for a range of manipulation-related tasks.
Design, implement, and deliver approaches that bridge exploration and production readiness with imitation learning, diffusion policy, VLA models, or representation learning.
Build data pipelines, training scripts, evaluation metrics, and experiment reports.
Analyze successful and failed robot trials to identify learnable patterns and production-relevant failure modes.
Collaborate with planning, perception, and controls engineers to understand how learned models could augment the production system.
Present clear technical findings, including what worked, what failed, and what should be tested next.
Required Skills
Current MS or PhD student in robotics or machine learning.
Experience with robot learning, imitation learning, reinforcement learning, diffusion policies, VLA models, or visuomotor policy learning.
Strong Python programming skills and hands-on experience with PyTorch, JAX, or similar ML frameworks.
Strong data debugging skills, including dataset inspection, failure analysis, and metric design.
Familiarity with ROS, ROS2, robot logs, RGB-D data, point clouds, or robot kinematics.
Comfortable reading papers, implementing baselines, running experiments, and communicating technical tradeoffs.
Nice to Have
Experience working with real robot data, not only simulation.
Prior work on manipulation, grasping, or mobile manipulation robotics.
Experience with large pretrained models, vision-language models, or multimodal representation learning.
Publications, open-source projects, or strong course/research projects in robot learning.
What Success Looks Like
By the end of the internship, the intern should deliver a clear technical result: a trained model, reproducible experiment pipeline, evaluation report, and recommendation on whether the explored method is promising for Anyware’s production manipulation roadmap.
The goal is not just a demo. The goal is to deliver a validated model or method that demonstrates measurable improvement on defined manipulation tasks, along with clear evaluation results and a concrete recommendation for integration into the production system.
Why Anyware
An A Team: Anyware is built by top talents. Collectively, our team has won six Best Paper Awards and Finalists, including the ICRA 2025 Best Paper Award and ICRA 2024 Best Paper Award.
An A Robot: Our robot, Pixmo, won the Best Innovation Award (top-1) at both ProMat 2025 and MODEX 2026 — the two largest logistics automation exhibitions in the world. This is the first time in history that a robotics company has won this prestigious award back-to-back. Pixmo also won the RBR50 Robotics Innovation Award, recognizing it as one of the top 50 most innovative robotics solutions globally.
Real Impact: You will contribute directly to the core manipulation intelligence that powers every Pixmo deployment. Your work does not stop at a demo — it ships into real customer sites and solves real problems at scale.
Fast Growth: You will sharpen your skills with strong ownership and real responsibility. The team size is doubling every year, creating plenty of room for talents to step up and shine.
Compensation & Benefits
The expected salary range for this role is $30-60/hr, depending on experience and qualifications.
Benefits include daily meal, per diem, and unlimited snacks and beverages in the office.
Equal Opportunity Statement
Anyware is an equal opportunity employer. We are committed to creating an inclusive workplace and welcome applicants from all backgrounds. Employment decisions are based on qualifications, merit, and business needs.
How To Apply
Please submit your resume and any relevant materials through the Anyware Career platform. We look forward to learning more about your experience and interest in joining our team.
Salary : $30 - $60