Compare 3 humanoids
UC Berkeley Humanoid Lite, Stanford University ToddlerBot, Weave Robotics Isaac 1 compared on the things that decide a purchase: price, capability, what actually ships, and estimated costs to your destination.
Delivered price, head to head
Delivered estimates to United States for 2 of 3 machines. The remaining costs need confirmation.
UC Berkeley
Humanoid Lite
Not orderable
Research access only; confirm procurement directly.
Stanford University
ToddlerBot
$6,101
to $6,393 landed
Domestic, no import
Weave Robotics
Isaac 1
$8,100
to $8,392 landed
Domestic, no import
Delivered to United States: the import estimate (machine plus freight, Sales tax (0%), robot duty, and broker), or a seller in your own market where one lists the machine for less. Accessories excluded. Check configuration and request a seller quote on each robot's datasheet.
Some costs are unknown, so no overall lowest-cost machine can be identified.
Compare the details
20 rows, 16 with differences
Scroll within the table to see all details. Model names and row labels stay visible. Unknown specifications remain marked as not published.
| Specification | ![]() UC Berkeley Humanoid Lite | ![]() Stanford University ToddlerBot | ![]() Weave Robotics Isaac 1 |
|---|---|---|---|
| Priced configuration | See seller configuration | See seller configuration | See seller configuration |
| Listed price | $5,000 | $6,000 | $7,999 |
| Class | Dev / demo | Dev / demo | Dev / demo |
| Who can buy | Not for sale | Consumer | Consumer |
| Status | Research institutions | Shipping now | Preorder |
| Capability rating | 2/5 | 3/5 | 4/5 |
| Maturity rating | 2/5 | 5/5 | 3/5 |
| Origin | United States | United States | United States |
| Size class | Compact | Compact | Wheeled base |
| Height | 80 cm / 2 ft 7 in | 56 cm / 1 ft 10 in | 175 cm / 5 ft 9 in |
| Footprint (W x D) | 38 x 22 cm | Not published | 55 x 55 cm |
| Weight | 16 kg | 3.4 kg | Not published |
| Degrees of freedom | 22 | 30 | Not published |
| Battery (listed) | 30 min | 19 min | 480 min |
| Payload (listed) | Not published | 1.484 kg | Not published |
| Hands | No hands, none offered | Grippers, not dexterous hands | Grippers, not dexterous hands |
| Out of the box | None. After a successful build you get an RL-controlled walking research humanoid with a full sim-to-real pipeline. | Nothing autonomous on power-up. ToddlerBot is a research kit: out of the box you get a 3D-printed bipedal chassis driven by DYNAMIXEL servos plus a fully pip-installable Python codebase. Its headline feature is a high-fidelity digital twin with plug-and-play zero-point calibration and motor system identification, so a policy trained in MuJoCo simulation transfers zero-shot to the real robot. After assembly and calibration you can run the provided policies (walking at up to 0.25 m/s, push-ups, pull-ups, wagon pushing, bimanual and full-body manipulation) or teleoperate it with a handheld gaming PC, but there is no app, voice assistant, or ready-made behavior for a non-programmer. | Out of the box it is a consumer appliance, not a kit: after it maps your home, it performs two scheduled routines with zero programming. "Laundry Flow" (finds dirty clothes, folds and puts away clean ones) and "Daily Reset" (makes the bed, fluffs pillows, tidies shoes, toys, and clutter into place). You control and schedule it from a smartphone app. When it hits a task it cannot do autonomously, a remote Weave operator briefly takes over to finish the job. |
| Development access | Everything open source under permissive licenses: hardware design, embedded code, and Isaac-based RL frameworks. The reference DIY research humanoid in the US. | Entirely open-source Python stack (hshi74/toddlerbot on GitHub). Codebase is MIT-licensed (commercial use allowed) and covers low-level control, RL training, diffusion-policy training, and real-world deployment; built on MuJoCo with MJX for accelerated training. The mechanical design files (Onshape, STLs) are CC BY-NC-SA 4.0, so the hardware design is non-commercial only. Full API docs, assembly manual, and bill of materials are published. | No public SDK, API, or EDU/developer program has been announced. Isaac 1 is a closed consumer appliance whose behaviors (Laundry Flow, Daily Reset) are defined by Weave and delivered through the app and cloud service. Buyers cannot write custom tasks or access low-level control. |
| Best for | Well-equipped labs and determined hobbyists who want a from-scratch, fully open RL research humanoid. | Robotics and embodied-AI researchers, university labs, and serious makers who want an affordable, reproducible, fully open bipedal platform to develop and test sim-to-real reinforcement-learning and imitation-learning policies (loco-manipulation), not buyers wanting a finished consumer robot. | US (initially California) early-adopter households who want a real, orderable home chore robot and accept a hybrid autonomous/teleoperated appliance rather than a fully independent or programmable humanoid. |
| Listing checked | 2026-09: sources for UC Berkeley Humanoid Lite | 2026-09: sources for Stanford University ToddlerBot | 2026-09: sources for Weave Robotics Isaac 1 |
Listed prices may cover different configurations and tax terms. Use the delivered estimates above for your destination. Battery and payload figures are manufacturer specifications, not results from a shared test.
Humanoid Lite
In the box
- Nothing ships: open-source repo with CAD, BOM, firmware, and an RL training/deployment stack
Not included
- The robot (DIY: 3D printing + assembly)
- Support/warranty
- Out-of-box demos
ToddlerBot
In the box
- ROBOTIS Bundle Kit ($4,335.51): the full set of DYNAMIXEL smart servos that actuate the robot (2x 2XC430-W250-T, 6x 2XL430-W250-T, 8x XC330-T288-T, 4x XC430-T240BB-T, 4x XM430-W210-T), plus HN11-I101 and HN12-I101 horn/bracket sets, a U2D2 USB communication controller and U2D2 PHB power set
- Alternatively, WowRobo ToddlerBot 2.0 assembled version ($4,299): all structural components, servos, and sensors, assembled and tested as a chassis
Not included
- NVIDIA Jetson Orin NX 16GB onboard computer (must be purchased separately, this is ~half the cost of a full build)
- Lithium battery pack
- WiFi module
- The 3D-printed structural shells and end-effectors (the ROBOTIS kit is DYNAMIXEL parts ONLY; you 3D print the body yourself from the open Onshape/STL files, ~21 min print time per repair part)
- Cameras, IMU, speaker, and microphones if buying the bare ROBOTIS parts kit
- A teleoperation controller (a Steam Deck or ROG Ally X handheld PC used to drive it)
- Any autonomy: it ships as hardware plus open-source code, not a robot that does anything on power-up
- Commercial-use rights to the mechanical design (design files are CC BY-NC-SA 4.0, non-commercial; only the software is MIT-licensed)
- Vendor technical support or a warranty in the consumer-product sense; support is a research community (Discord/WeChat)
Isaac 1
In the box
- Isaac 1 wheeled home robot (adjustable-height torso on a motorized wheeled base)
- Two arms with two-finger grippers (orange-tipped claws)
- Onboard cameras and AI perception for autonomous operation
- Charging dock / recharge capability (robot self-recharges between shifts)
- Smartphone companion app for control and scheduling
- Hardware camera privacy controls (physical camera disable)
Not included
- Human-like hands or dexterous fingers (it uses simple two-finger grippers, not fingered hands)
- A washing machine operator: Isaac 1 folds and sorts laundry but does not run the washer/dryer
- Full unsupervised autonomy: hard tasks fall back to a remote human teleoperator
- A developer SDK or programmable API (this is a closed consumer appliance, not a dev/research platform)
- Legged walking or stairs capability (it is wheeled and cannot climb stairs)
- Ownership free of ongoing service: the experience is tied to Weave's cloud service and app
What the evidence tells you
- Highest capability rating
- Isaac 1: 4/5.
- Highest maturity rating
- ToddlerBot: 5/5.
- Lowest listed weight
- 2 of 3 values are available. A leader cannot be established.
Capability and maturity are editorial ratings based on specifications and sales availability. They do not measure reliability, autonomy, or performance on your task. Each datasheet shows the scoring factors.


