> For the complete documentation index, see [llms.txt](https://x.ancorasir.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://x.ancorasir.com/rob7103-8103/student.md). # \[ROB7103/8103] Embodied AI Systems @ MBZUAI 2026 Autumn ### Course Description This seven-week course examines how AI methods become parts of measurable, reproducible, and responsibly operated physical systems. It assumes prior competence in machine learning, computer vision, reinforcement learning, and related AI methods. It does not repeat general AI, kinematics, control, or planning courses. Instead, it concentrates on embodiment, sensing, actuation, communication, calibration, fabrication, simulation, Sim2Real evidence, robot-learning deployment, and experimental diagnosis. The course normally contains three scheduled events per week, with durations determined by the official timetable. It includes six linked Embodied AI Meets Design worksheets, one revised worksheet portfolio, and one team final project. An optional two-day Seeed hackathon may be offered as a separate supervised event. Details will be announced separately. The course begins on October 20, 2026. Access to the supporting source repositories is managed separately and does not determine the copyright or licensing status of the rendered materials. ### Course Aims The course aims to enable students to design, build, instrument, and critically evaluate embodied AI systems. It bridges advanced AI knowledge and physical implementation without repeating foundational AI, kinematics, planning, or control theory. Students learn to translate a bounded system idea into a feasible physical-digital prototype, establish measurement validity, integrate a learning-enabled component, and evaluate the resulting capability with appropriate evidence. The course emphasizes data semantics and provenance, calibration, frames, timing, Sim2Real discrepancies, safety, reproducibility, and failure analysis. By the end of the course, students should be able to develop a bounded team prototype, compare learned and non-learned baselines, communicate limitations, and identify a defensible next step. The course also develops practical competence in CAD, limited rapid prototyping, simulation, robot interfaces, experiment design, and technical communication. ### Learning Outcomes By the end of the course, a successful student should be able to: 1. Define and critique an embodied-AI system. 2. Engineer and bring a physical–digital platform into operation. 3. Produce research-quality embodied experience. 4. Evaluate correspondence between computational and physical representations. 5. Assess learning inside a real robotic feedback cycle. 6. Deliver and defend a repeatable team prototype. ### Content Summary The course introduces embodied AI as a closed-loop physical system and uses an iterative project process to connect theory, system design, physical experimentation, learning, and responsible operation. Topics may include physical and actuator modeling; parameter identification; CAD, fabrication, and introductory 3D printing; robot descriptions and simulation; hardware commissioning and calibration; multi-actuator communication; robot morphology; multimodal sensing; frames, clocks, and data interfaces; motion retargeting and demonstration data; vision-guided manipulation; imitation learning; Sim2Real validation; digital models and digital twins; experimental baselines; uncertainty and failure analysis; safety and human authority; reproducible code and data lineage; and evidence-supported project decisions. Examples may use single- or multi-actuator systems, manipulation platforms, mobile perception, simple robot arms, or combinations of compatible systems. Six concise project worksheets guide the progressive development, evaluation, revision, and final assessment of a reproducible physical prototype. Specific frameworks, tools, systems, and open-source resources are selected according to course needs and may change between offerings. ### Assumed Knowledge Students are expected to have graduate-level programming ability in Python and experience using Linux, command-line tools, and version control. They should understand the fundamentals of machine learning and computer vision and have prior exposure to reinforcement learning or robot learning. Working knowledge of linear algebra, calculus, probability, and experimental data analysis is expected. Familiarity with basic robotics concepts, coordinate frames, kinematics, dynamics, control, and planning is helpful because these topics are applied but not retaught systematically. Students must be willing to work collaboratively and follow laboratory, electrical, mechanical, data, and robot-safety instructions. ### Course Instructor and Teaching Team * **Lead instructor:** Prof. SONG Chaoyang · Associate Professor of Robotics, MBZUAI * **Email:** Chaoyang.Song \[at] mbzuai.ac.ae * **Office:** C-1.01, Building 1B · **Office hours:** Mondays, 10:00–12:00 * **Teaching Assistants** * ZHANG Dunxing (Dunxing.Zhang \[at] mbzuai.ac.ae) * YANG Hang (Hang.Yang \[at] mbzuai.ac.ae) * **Technical Support** * Dr. Mohamed HALWANI (Mohamed.Halwani \[at] mbzuai.ac.ae) * WEI Jinqi (Jinqi.Wei \[at] mbzuai.ac.ae) * LI Zhibin (Zhibin.Li \[at] mbzuai.ac.ae) * **Research Support**: Prof. WAN Fang (wanf \[at] sustech.edu.cn) ### Grading Policy Assessment combines individual participation with a progressive team project. * Individual (10%) * 10%: Recorded attendance across scheduled meetings * Team (90%) * 30%: Six staged project records * 25%: Working system and reproducibility package * 10%: Consolidated and updated project record * 10%: Narrated project video * 10%: Oral briefing and response to questions * 5%: Technical poster ### Academic Integrity All work must comply with the University's academic-integrity requirements and with the specific collaboration and tool-use conditions stated for each assessment. Because this course depends on shared code, research papers, datasets, pretrained models, CAD files, and external software, clear attribution and traceability are essential parts of scholarly practice. Students are expected to: * cite publications, documentation, datasets, software, models, hardware designs, media, and other reused material; * preserve license notices and observe any restrictions attached to third-party resources; * distinguish their own contribution from instructor-provided, teammate-produced, open-source, and machine-generated material; * maintain authentic experiment records, including unsuccessful trials and relevant changes to hardware, code, parameters, or data; * verify technical claims and inspect generated code or text before including it in assessed work; * disclose the use of generative AI or automated coding tools when an assessment brief requires it; and * retain sufficient development history to explain how the submitted result was produced. Examples of unacceptable practice include inventing or altering measurements, hiding failed runs that materially affect a conclusion, presenting another person's work as one's own, submitting unverified machine-generated content, disguising the source of reused assets, sharing restricted assessment material, or bypassing safety and authorization controls to obtain a result. ### Recommended Textbook(s) 1. Featherstone, R. **Rigid Body Dynamics Algorithms**. Springer, 2008. ISBN 978-0-387-74314-1. 2. Siciliano, B., and Khatib, O., editors. **Springer Handbook of Robotics**. 2nd edition. Springer, 2016. ISBN 978-3-319-32550-7. 3. Lynch, K. M., and Park, F. C. **Modern Robotics: Mechanics, Planning, and Control**. Cambridge University Press, 2017. ISBN 978-1-107-15630-2. 4. Sutton, R. S., and Barto, A. G. **Reinforcement Learning: An Introduction**. 2nd edition. MIT Press, 2018. ISBN 978-0-262-03924-6. ### Teaching Schedule
WkTue 0900-1020 CR6Wed 1030-1150 CR6Fri 0900-1050 L12
01

Oct 20: Class 01

Course Introduction with asMagicBrain

Oct 21: Class 02

Actuator Basics for Controlled Action

Oct 23: Class 03

Basic Digitization of Physical Embodiment

02

Oct 27: Class 04

Basic Quantification of Sim2Real Gap

Oct 28: Class 05

From Single-Actuator to Multi-Actuator

Oct 30: Class 06

Learned In-Hand Cube Reorientation

03

Nov 03: Class 07

Rotating a Cube with Visual Feedback

Nov 04: Class 08

Let's Build a Simple Arm & Turn it On

Nov 06: Class 09

Teleoperation by Streaming from asMagic

04

Nov 10: Class 10

Collect Data for Reinforcement Learning

Nov 11: Class 11

Sim2Real Gap in Hardware Deployment

Nov 13: Class 12

Final Project I: Claim

05

Nov 17: Class 13

Final Project II: Experience

Nov 18: Class 14

Final Project III: System

Nov 20: Class 15

Final Project IV: Grounding

06

Nov 24: Class 16

Final Project V: Learning

Nov 25: Class 17

Final Project VI: Decision

Nov 27: Class 18

Final Poster Submission

07

Dec 01: Class 19

Final Code Submission

(No Class. Public Holiday)

Dec 02: Class 20

Final Video Submission

(No Class. Public Holiday)

Dec 04: Class 21

Final Project Showcase

### Important Deadlines * Nov 13 @ 1050: Worksheet 1 — Claim * Nov 17 @ 1020: Worksheet 2 — Experience * Nov 18 @ 1150: Worksheet 3 — System * Nov 20 @ 1050: Worksheet 4 — Grounding * Nov 24 @ 1020: Worksheet 5 — Learning * Nov 25 @ 1150: Worksheet 6 — Decision * Nov 27 @ 1050: Final Poster * Dec 01 @ 1020: Final Code * Dec 02 @ 1150: Final Video * Dec 04 @ 0900: Final Presentation & Final Demonstration ### How to Prepare Bring a laptop and prepare to use Python, command-line tools, and Git. Use isolated virtual environments for class software. Follow the instructions supplied with each exercise and record the software versions used. The course combines technical explanations, worked demonstrations and practical activities. Review the supplied material before each session. Retain code, measurements, and experiment notes so another team can reproduce your results. Using asMagicBrain is optional. You can access course materials and project work through GitHub and standard development tools. Course code and downloadable materials are available in the [Students repository](https://github.com/asEmbodiedAISyst/2026Autumn). ### Final Project and Worksheets Develop one team project through six worksheets: **Claim → Experience → System → Grounding → Learning → Decision**. The team guide and worksheet files will be provided before the project phase. Complete one worksheet for each of Classes 12–17. Submit it handwritten and signed by every team member before the end of the corresponding session. You may revise earlier worksheets as your project develops; keep earlier submissions and explain any material changes. The final project package includes the worksheet portfolio, reproducible code and environment instructions, relevant data and results, a technical poster and a narrated video. Present and demonstrate the project at the final showcase. Detailed artifact requirements will be supplied separately. Each team will have a private project repository for its instructions, development and final submission. We will provide repository links when teams are assigned. ### Course Materials This page provides the Autumn 2026 course information and schedule. Individual class pages are not available yet. Dates and assessment details follow the current course plan; we will announce any approved changes here. No-class dates in the schedule do not automatically cancel listed submission deadlines. All times are Abu Dhabi time (UTC+4). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://x.ancorasir.com/rob7103-8103/student.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.