A lab does not want a product. It wants a body it can attach a policy to, with the low-level control open enough that the research question is not blocked by the vendor's stack, and a simulation model that matches the hardware closely enough to be worth training in.
The other half of the requirement is unglamorous: parts. A platform students cannot repair themselves stops being a research platform the first term something snaps.
Requirements are stated with the reason they exist. A requirement without a reason is a preference, and preferences are how a shortlist quietly becomes a recommendation.
A high-level walk command is a demo. Research needs torque or position at the joint.
Policies are trained in simulation and transferred; without a matching model the platform is a very slow simulator.
It is what the students already know and what the lab's existing stack is written against.
Hardware gets broken deliberately here. Whole-unit RMA is not a spare-parts policy.
A grant buys one platform. Price decides this category more than any specification.
These are not weaknesses to trade off against a high score. Any one of them ends the conversation, which is why they sit outside the scoring entirely.
Shipping robots in our catalogue whose class matches this job, listed under the research weighting so you can scan them — not a recommendation. The buyer's success is measured in papers and graduates, so openness and expandability are the product.
Order follows the research profile score. Requirements and disqualifiers above are editorial; they do not remove robots from this list. How the score works
Compare candidatesThese are kept off the candidate list rather than mixed into it. A candidate list is something you can act on, and hardware nobody can order does not belong on one.
NEURA Robotics
NEURA Robotics
Hexagon
AKINROBOTICS
Boston Dynamics
Xiaomi
Dobot
Holiday Robotics
Figure AI
Figure AI
Generative Bionics
Humanoid
XPENG Robotics
Weave Robotics
PAL Robotics
KEENON Robotics
KEENON Robotics
Leju Robotics
Kawasaki Robotics
Robot Era
Matrix Robotics
Matrix Robotics
Mentee Robotics
1X Technologies
Tesla
EngineAI
Pudu Robotics
Pudu Robotics
Pudu Robotics
UniX AI
Sanctuary AI
PrimeBOT
PrimeBOT
LimX Dynamics
Tokyo Robotics
UBTECH Robotics
Well served, and the reason the cheaper platforms have taken the field: a lab that can buy three of something and break one is more productive than a lab with a single flagship it is afraid to use. Capability per unit is now the wrong metric here.
