Stanford University · ME Dept
ARM Lab
ARM Lab
Member Handbook
Assistive Robotics and Manipulation Laboratory
Assistive Robotics and Manipulation Laboratory
This document is for ARM Lab members only. Contact the lab administrator if you need access.
Policies, procedures, and resources for everyone in the Assistive Robotics and Manipulation Laboratory. Read this thoroughly when you join — and revisit it whenever you have questions about how we work.
We're excited to have you here. The ARM Lab works at the intersection of robotics, manipulation, and human-machine systems. Our goal is to do rigorous, impactful research in a collaborative and supportive environment.
This handbook covers everything you need to know to get started: how to get physical access, how we handle authorship and publications, what's expected of different researcher roles, and how to take care of the space we all share.
Before you can work in the lab, you need to complete safety training and get keycard access. Work through these steps in order:
As of Winter 2025, the School of Engineering requires that all student researchers be either (a) enrolled in research course units, or (b) have a paid RAship from a fellowship or grant. Each quarter, every researcher must meet one of these conditions.
If enrollment would exceed your unit cap, we will unfortunately be unable to collaborate that quarter. Plan ahead — don't pay the extra-unit fee.
MS students who expect to work in the lab for 3–6 quarters should pace themselves: take only 1–2 units per quarter (ME 391/392) to make the 12-unit research cap last. Beyond the cap, MS students may pursue a "distinction in research" (requires Prof. Kennedy's approval and must begin in the first year). Peer-reviewed publications will carry equal or greater weight than an MS thesis for PhD applications.
Undergrads should reserve at least 1 unit per quarter in their academic plan for research (ME 191/192). Note this typically does not count toward degree requirements, but publications and letters of recommendation are valuable for graduate school and technical careers.
The ARM Lab follows the IEEE Ethical Requirements definition of authorship. To be listed as an author, all three criteria must be met:
Contributors who do not meet all three criteria may be included in the Acknowledgments section instead.
Before submitting, ask yourself:
Collaboration within the lab is free-flowing. Students have broad freedom to explore ideas and work together — but all new research projects or directions must be communicated to Prof. Kennedy first. This ensures advocacy for all parties, technical guidance, and connection to additional resources (people, equipment, space).
Cross-lab collaborations are generally encouraged, but every PI must ensure student effort aligns with their grants and research goals. Consult Prof. Kennedy before moving forward with any inter-lab collaboration.
External collaborations require the most care:
In all collaboration contexts — internal, Stanford-wide, or external — Prof. Kennedy must be consulted and will guide the process from initiation through ongoing maintenance.
The typical ARM Lab PhD is 5–6 years. Prior coursework (e.g. from an MS) can accelerate the timeline if applied by ME Student Services. To defend, students must meet the minimum publication bar:
Minimum to graduate before year 5: 5 conference papers + at least 1 journal publication, all in strong impact-factor venues, all as (co-)first author. For each year beyond year 5, one additional conference or journal paper is expected. More is always encouraged.
Alternative publication profiles are acceptable — for example, 3 journals + 2–3 conference papers. What matters is quality venue and (co-)first authorship.
Lab-internal and Stanford-internal collaborations are allowed and encouraged, with Prof. Kennedy's validation. External (especially industry) collaborations require IP agreements and Prof. Kennedy's sign-off before engaging lab resources.
Discuss internship timing and funding implications with Prof. Kennedy before committing.
Make sure you have access to all of the following before you start working:
If you don't have access to any of these, ask a current lab member or contact Max Burns.
The lab maintains a shared pool of equipment. Some devices have dedicated Slack channels — check there first. For passwords, setup guidance, and resources, ask a current lab member who has used the specific device.
This list may not be complete — see the Lab Inventory tool (below) for the full, searchable catalog.
To conduct a non-medical study with human participants, you must complete IRB training and be added to the lab protocol. Both steps are required before running any participants.
Do not begin data collection until you have received IRB approval. Protocol modifications can take time — plan accordingly.
One major contribution in many of our works is the publication and open-sourcing of our datasets and code. For code, we use GitHub (github.com/armlabstanford) ↗ and hold our code to an industrial standard of quality for wide adoption and use — clean APIs, documentation, and reproducible environments.
For datasets, we prioritize accessible, free platforms with long-term stability. The right choice depends on your dataset's size, format, and intended audience. The main options are:
Stanford Libraries' official long-term preservation platform. Assigns DOIs, integrates with Stanford's research infrastructure, and carries institutional credibility. Best for datasets you want formally archived under Stanford's stewardship. Contact the library for size and access options.
If your dataset is ML-related, this is increasingly the community standard. Free, no practical size limit for public datasets, good versioning with Git LFS, and a large active community. Excellent discoverability for robotics and perception work.
Designed specifically for large research datasets — peer-to-peer, so the more people download, the faster it gets. No storage limit, completely free, and used routinely for multi-TB ML datasets. The tradeoff: users need a torrent client.
Backed by CERN, free, and assigns DOIs automatically. Default limit is 50 GB per record, but quota increases are available for legitimate research use — they're generally accommodating. Good all-around option for moderately sized datasets that need a citable DOI.
Supports up to 300 GB free, making it suitable for large datasets that don't need a torrent setup. Worth checking whether your target journals or field communities recommend it — some mandate Dryad for data availability requirements.
Whichever platform you choose, discuss it with Prof. Kennedy before publishing. Some funding sources have data-sharing requirements that may constrain your options, and the choice of platform should be made early — not after the paper is accepted.
The ARM Lab is a shared, professional research environment. Keeping it clean, organized, and ready for the next person is everyone's responsibility — not a secondary concern. These aren't arbitrary rules; a well-maintained lab reduces errors, protects expensive equipment, and makes everyone's work go more smoothly.
Dispose of all food packaging immediately after eating. No open containers, wrappers, or crumbs left on benches, desks, or near equipment.
After working at a shared desk or bench, return it to the state you found it. Quick wipe-downs go a long way.
Cables, tools, and adapters should go back to their designated location after each session — not left on the bench "for later."
Don't leave waste at your workspace. If the bin is full, take it out or notify a lab member who can.
When you finish a session with shared equipment, apply the following checklist before you leave:
The rule of thumb: leave the lab in a state you'd be happy to find it in when you arrive for an early morning session.
The ARM Lab maintains a searchable inventory of all equipment, sensors, and lab items. Before purchasing something new or searching the physical space for a component, check the inventory first.
Searchable catalog of all ARM Lab equipment and components.
If you find an item in the lab that is not in the inventory — a component with no label, an unknown sensor, a mystery cable — do not discard it or move it without checking first. Instead:
Keeping the inventory accurate and complete is a collective effort. When you bring new equipment into the lab — from a purchase, a loan, or a project — add it to the inventory before using it.