PHYSICAL AI EDUCATION · SV ROBOTICS
Physical AI classes, ready from day one
A 5-axis robot arm and a depth camera on a mobile wheeled robot. Students train AI in a virtual world, then load that AI onto the real robot to find, pick and move blocks. The robot ships with a complete 20-session course — textbook, teacher's guide, teacher training and remote technical support — so a teacher new to the subject can lead the very first class.
10 foundation + 10 advanced
Two tracks for two levels
Every student operates a robot
Imitation learning · reinforcement learning · vision-language-action
WHY AMR ROBOT ARM
It maps and drives with lidar, finds objects with the depth camera and picks them up with the arm. Schools teach the core skills of Physical AI with one robot instead of buying several.
Students train AI with thousands of virtual robots in Isaac Sim and Isaac Lab, then put it on the real robot — the same Sim-to-Real workflow research labs use.
Textbook, teacher's guide, lab code and quizzes for every session, teacher training before classes, and remote technical support on weekdays during classes.
FOR PARTNERS
A Physical AI education robot: a mobile wheeled robot carrying a 5-axis robot arm, a depth camera and lidar. It comes with a 20-session course for high schools and universities in which students train AI in a virtual world and deploy it to the real robot.
The robots, GPU PCs for Sim-to-Real practice (required for universities, optional for high schools), a student textbook and a teacher's guide, lab code for every lesson, per-session quizzes, teacher training, and remote technical support on weekdays from 09:00 to 18:00 KST.
Not for the high school course: students work with trained models, no GPU required. With GPU PCs, students also train AI themselves on virtual robots. The university course trains directly on GPU PCs.
Two students per robot. The number of robots is set by class size, so every student operates a robot.
20 sessions: 10 foundation and 10 advanced, about 3 hours each, scheduled to fit the school timetable.
NVIDIA Isaac Sim and Isaac Lab for the virtual world and training, and ROS 2 on the robot. Some AI lessons use the internet and external AI services.
Per-session quizzes, the execution logs the robot records, and the final mission presentation. Students repeat hand-coded and AI methods under the same conditions and explain the results with a comparison table.
PRODUCT
Five moving joints and a gripper pick up and move objects such as blocks.
Mounted on the arm, so the camera looks wherever the arm points.
Drives forward, backward, turns, and also slides sideways.
Measures distances around the robot for mapping, navigation and obstacle avoidance.
HOW STUDENTS LEARN
By demonstration, by reward, by language — students teach the same block-picking task three ways and compare the results.
A person moves the arm by hand to demonstrate; the robot learns to pick by copying the demonstrations.
Sessions 8 · 12 · 13The robot learns by itself in a virtual world while students change the reward, then the result goes onto the real robot (Sim-to-Real).
Sessions 9 · 14 · 15 · 16An AI that understands spoken instructions and acts on them is fine-tuned for our robot, and the three methods are compared.
Sessions 18 · 19High school / university. High schools start without GPUs using trained models; universities train directly on GPU PCs.
Sessions 11, 17, 18 and 19 (language-based targeting, language models, VLA ① ②) are chosen to fit the school's level and equipment.
About 3 hours per session, scheduled to fit the timetable.
Design → test in simulation → deploy on the real robot → compare with data
20 SESSIONS
Both tracks have 10 foundation and 10 advanced sessions and follow the same sequence at different depths. Each session is about 3 hours.
Teachers practice with the robot and the virtual world first
ROS 2 · arm coordinates · depth camera · lidar mapping and navigation · virtual world
Imitation learning · RL and Sim-to-Real · language models · VLA
Teams design, demo and present a tidy-up robot you instruct by voice
Foundation · Sessions 1–10
Watch the finished tidy-up robot demo; safety rules
Remote control; inspect topics
Move the arm with joint and end-point coordinates
Measure the distance to an object
Map the classroom
Drive to a goal on the map autonomously
Control the robot in the virtual world
Run and observe an arm that copies a demonstration
Compare results across rewards; deploy to the real robot
Find blocks with a provided model
Advanced · Sessions 11–20
Compare recognition results across sentences
Record demonstrations by hand-guiding the arm
Pick autonomously with a trained model
Compare results across randomization ranges
Avoid real obstacles; compare with the virtual run
Record success counts; discuss causes
Change the command and check the plan
Observe the motion for each command
Compare models before and after fine-tuning
Combine the skills learned to complete and present the mission
Optional: sessions the school includes or runs as a demo, depending on level. Trained model: sessions that use a trained RL model so students can practice without a GPU.
Foundation · Sessions 1–10
Technology overview; safety rules
Write nodes and topics
Implement forward and inverse kinematics
Implement pixel-to-3D conversion
Build a SLAM map and tune settings
Tune path planning; analyze driving logs
Modify the robot model and environment
Behavior cloning principles; demonstration data structure
Reward design and training experiments; measure real-world error
Small-scale extra training; measure accuracy
Advanced · Sessions 11–20
Analyze failure cases; prompt design
Collect and organize demonstration data
Compare results by number of demonstrations
Set up randomization and train
Train the avoidance policy and deploy it
Classify failure types and report
Design functions; implement tool calling
Set up the inference server and client
Fine-tune on the class's own demonstration data
Experiments to justify each technology choice; short report and presentation
In the IL, RL and VLA sessions of the university course, students train on GPU PCs themselves and benchmark against a reference model.
SIM-TO-REAL
Students go through this workflow in session 9 and sessions 14–16.
Set up a robot and blocks in Isaac Sim that match the real ones
Thousands of virtual robots practice at the same time in Isaac Lab
Export what the robot learned in the virtual world as a model
Load it onto the real AMR Robot Arm and close the gap between virtual and real
Check that the real robot completes the same task
High school
University
FINAL MISSION · SESSION 20
“Put the red block in the blue box.”
Teams combine everything from the semester into one project. A student gives an instruction; the robot understands it, drives to the block, picks it up and moves it.
The AI understands the instruction and decides what to do
Finds the block with the camera and drives to it
Picks it up with the chosen method: IL, RL or VLA
Carries it to the box and drops it in
AI vision (sessions 10–11), language-model planning (session 17), lidar navigation (session 6) and the chosen picking method come together; each team designs, demos and presents.
ASSESSMENT
Check understanding of the key concepts.
Review the logs and results the robot records.
Design, demo and presentation are assessed together.
Hand-coded AI (reinforcement learning) Mean of 3 runs, range in brackets · shorter is faster
Whole task (incl. stages not shown) 63.3 s → 56.3 s · about 88% of the 7.0 s gap comes from Approach
| Stage | Hand-coded | AI (RL) | Reading |
|---|---|---|---|
| Find | 6.0 s | 6.2 s | Same code → no difference |
| Approach | 19.4 s (18.4–19.9) | 13.2 s (13.1–13.3) | AI used here; ranges do not overlap → real effect |
| Align | 11.4 s (9.4–13.6) | 10.3 s (9.7–11.1) | Ranges overlap → not claimed as an effect |
| Whole task (incl. stages not shown) | 63.3 s | 56.3 s | About 88% of the 7.0 s gap comes from Approach |
OUTCOMES
Connect seeing, thinking and moving in code
Teach the same task with IL, RL and VLA, and compare
Find and close the gaps between simulation and reality (Sim-to-Real)
Explain hand-coded versus AI methods with measured data
WHAT YOU GET
Not just a robot. Every session is built for hands-on practice on the real AMR Robot Arm, with textbook, code and assessment included.
Two students per robot, so every student operates one.
Required for universities, optional for high schools. Sized to the school.
Teaching notes and tips for every session.
Ready-to-run code for each lesson in the textbook.
Check understanding of the key concepts every session.
Training before classes; remote support on weekdays, 09:00–18:00 KST.
Uses NVIDIA Isaac Sim and Isaac Lab. Some AI lessons use the internet and external AI services.
HOW TO START
Teachers and instructors practice with the robot and the virtual world before classes begin. The teacher's guide has teaching notes and tips for every session.
Remote support on weekdays, 09:00–18:00 KST
Tell us the level (high school or university), number of students and whether GPU PCs are available, and we propose the number of robots and a course plan.
PARTNER WITH US
Tell us your country or region, the institutions you work with (high schools, universities) and the expected number of students. We will reply with a proposed robot and course configuration. The course guides below are in Korean.
20-session curriculum, starting without GPUs, a sample student comparison table, getting started
DownloadPDF · 4 pages · EnglishUniversity course guide20-session curriculum with technical terms per session, GPU PC training labs, assessment, getting started
DownloadSV ROBOTICS · CONTACT
Talk to us about AMR Robot Arm.
Tell us your country or region, target institutions and timeline. Our team will reply by email.
Prefer email? Write to innorobo@naver.com.
