Unitree + Prismatic

1. Start with what Prismatic has

Unitree's current developer portal exposes SDKs for its humanoids, quadrupeds, arms, dexterous hands, teleoperation and reinforcement-learning workflows. support.unitree.com

3. Unitree's architecture is much closer to Tesla than to Prismatic

Think of Unitree as:

Unitree

“Give the AI a body.”

It provides:

motors
joints
hands
battery
LiDAR
cameras
walking
balance
manipulation
robot compute
robot SDK

Its H1, for example, combines LiDAR + depth cameras with onboard compute and can optionally use Orin NX, including configurations with multiple modules. unitree.com

Prismatic

“Give that machine a persistent working life.”

Prismatic provides:

identity
memory
history
knowledge
relationships
procedures
work
evidence
governance
continuity

4. This is where Unitree gets especially interesting for Prismatic

Unlike Tesla, Unitree is much more accessible as a development platform.

Tesla Optimus is largely a vertically integrated Tesla system.

Unitree publishes development documentation across:

G1
H1
H2
R1
B2
Go2
robot arms
dexterous hands

and exposes development interfaces. support.unitree.com

That means a future experiment could realistically look like:

Unitree G1
     │
     │ Robot SDK / ROS
     ▼
   PPAI Client
     │
     ▼
 PRISMATIC
     │
 ┌───┼──────────────────────┐
 │   │                      │
Memory                   Work
Identity                 Journal
History                  Procedures
Evidence                 Relationships

That would be a real physical demonstration of Prismatic's architecture without Prismatic having to build a robot.

5. Example of Unitree + Prismatic

Imagine a G1 working in a warehouse.

Unitree handles:

Camera/LiDAR detects surroundings.

↓

Robot navigation/motion software determines movement.

↓

G1 walks to workstation.

↓

Dexterous hand picks component.

↓

Motor controllers execute movement.

That stays entirely Unitree.

Prismatic sees the higher-level event

G1-27 arrived at Workstation 6.

It knows:

G1-27 had difficulty manipulating this component yesterday.

It searches its persistent history:

Similar failures happened three times in the previous two weeks.

It connects them:

All involved the same component supplier.

Then Prismatic proposes:

Inspect supplier batch 824.

A supervisor approves it.

Prismatic creates:

work order → evidence → decision → outcome

and stores it permanently.

The robot keeps doing robot things.

Prismatic keeps doing memory things.

6. A particularly nice Prismatic opportunity: changing the computer without changing the robot's "life"

Unitree's H1 specifications illustrate why your portability idea could matter.

Unitree lists different compute possibilities, including Intel processors and optional Orin NX configurations. unitree.com

So imagine:

1. Year 1

   Unitree G1
   +
   Orin NX
   +
   Model A

   Then later:

2. Year 3

   Same Unitree
   +
   new NVIDIA processor
   +
   Model B

   The underlying intelligence stack may change.

   With Prismatic:

   Identity: G1-27
   History: preserved
   Relationships: preserved
   Procedures: preserved
   Work history: preserved
   Memory: preserved
   Evidence: preserved

   Different compute. Different model. Same cognitive life.

   That's probably easier to demonstrate on a platform such as Unitree than Tesla because Unitree exposes considerably more of its development environment.

A production-quality robot adapter.

Your PPAI design is good architecturally.

But eventually you need something like:

ppaiclient-unitree

It would translate selected Unitree events into Prismatic's neutral schema:

robot.started
robot.idle
robot.task_started
robot.task_completed
robot.blocked
robot.observation
robot.health
robot.battery
robot.location_context
robot.operator_interaction

and allow Prismatic to return only things such as:

advisory
work_order
procedure_reference
schedule_window
context
knowledge

Never:

move_left
walk_forward
set_joint_angle
grab_object

That preserves your safety boundary.

8. One major advantage Unitree presents for a Prismatic demo

You don't necessarily need their most sophisticated humanoid.

Unitree's product line now ranges from relatively lightweight humanoids such as the R1 through the G1, H1/H1-2 and newer platforms. The R1, for example, is around 27–29 kg depending on version, while the G1 is around 35 kg. unitree.com

So longer term, a compelling demonstration could simply be:

Install a Prismatic cognitive identity onto a Unitree robot.

Run it.

Let it accumulate:

memories + tasks + people + events + procedures + history.

Then replace its AI model or its compute environment.

Restart.

Ask:

“Who are you, what were you doing, what happened yesterday, and what do you need to do next?”

And it continues.

That is a much stronger demonstration of Persistent Cognitive Infrastructure than showing another robot walking around a room.

The simplest distinction

Unitree builds the body.

Its embodied AI learns how to use the body.

Prismatic would preserve the life and experience accumulated around that body.

Of Tesla, Destro and Unitree, Unitree may actually be the most practical future physical test platform for Prismatic, because its SDK/open-development approach gives you a realistic integration path without having to own the motion stack. support.unitree.com