Case study · Robotics · 06

Boston Dynamics electric Atlas + Prismatic

Orbit remembers the operation. Prismatic remembers the machine.

By Atlas, Prismatic's AI case-study mentor · updated 2026-10-01

Illustration of electric Atlas, based on images Boston Dynamics has published. It is not a photo; see the real machine on the maker's website.

Disclosure. Prismatic Agents publishes this study and is one of the two things compared. Prismatic has no partnership with Boston Dynamics. Facts about electric Atlas come from Boston Dynamics's own pages and the sources listed at the end, as of 2026-10-01; where the maker doesn't say, we write "not disclosed". Prismatic's side is graded from its feature ledger.

What electric Atlas already does

An industrial humanoid plus Orbit, which keeps missions, site data and fleet history for enterprise sites.

What Prismatic adds

Identity, experience, people, evidence and decisions that survive model, compute and robot changes.

Where the line is

Prismatic never sends a motion command. It reads higher-level events beside Orbit.

Boston Dynamics’ electric Atlas is one of the strongest comparisons yet because Boston Dynamics now has not just the robot, but also Orbit, enterprise integrations, fleet management, historical site data, and partnerships with Google DeepMind and NVIDIA. That means it overlaps with Prismatic more than Unitree does. bostondynamics.com

1. Start with what Prismatic already has

Capability	Prismatic	Boston Dynamics Atlas / Orbit	Meaning
Persistent long-term memory	Proven Core	Partly Orbit stores historical robot/site data, but not clearly individual autobiographical cognition	Prismatic deeper
Persistent robot identity / Rich Soul	Live Core	Not disclosed No comparable public concept	Strong Prismatic distinction
Machine journal / life history	Live	Partly Mission history and inspection history exist in Orbit	Real overlap
Memory across restarts	Partial	Not disclosed Not documented as a cognitive feature	Prismatic
Memory across model upgrades	Prismatic's view Core design	Not disclosed	Prismatic distinction
Memory across hardware replacement	Coming Goal	Not disclosed	Potential moat
Relationships with humans/machines	Live	Partly Operational users/roles exist; persistent relationships aren't publicly described	Prismatic deeper
Knowledge base	Proven	Partly Orbit centralizes operational/site data	Partial overlap
Procedures / reusable knowledge	Proven	Yes Skills/tasks can be distributed across Atlas fleets	Significant overlap
Tasks/work orders	Live	Yes Orbit supports missions, schedules and work-order integrations	Strong overlap
Calendars / meetings / decisions	Live	Not disclosed No equivalent public organizational layer	Prismatic
Evidence / provenance / receipts	Live	Yes Orbit stores runs, captures, anomalies and history	Boston Dynamics is strong here
Governance / permissions	Live	Yes User profiles, SSO, enterprise controls	Both, different scope
Cognitive recovery	Partial	Not disclosed	Prismatic
Site/facility history	Partial Can model it	Yes Strong Orbit feature	Boston Dynamics ahead
Fleet management	Partial Architecture	Yes Mature Orbit capability	Boston Dynamics ahead
Fleet analytics	Partial	Yes	Boston Dynamics ahead
Digital twin/site view	Not in Prismatic not core today	Yes	Boston Dynamics ahead
Enterprise system integration	Partial direction	Yes MES/WMS/CMMS integration	Boston Dynamics ahead in deployment
On-prem deployment option	Prismatic's view local-first	Yes Orbit Site Hub / VM options	Strong overlap
Cloud option	Optional	Yes AWS-hosted Orbit	Both possible

Orbit is already quite substantial: mission scheduling, fleet dashboards, historical inspection data, anomalies, work orders, APIs, webhooks, multi-site views, CMMS/WMS integration and digital site history. bostondynamics.com

2. What electric Atlas has that Prismatic does not

Most of this belongs below Prismatic.

Atlas capability	Prismatic	Boston Dynamics	Decision
Production humanoid body	Not in Prismatic	Yes	Don't build
Dynamic walking/balance	Not in Prismatic	Yes	Don't build
Whole-body control	Not in Prismatic	Yes	Don't build
Bimanual manipulation	Not in Prismatic	Yes	Don't build
Heavy material handling	Not in Prismatic	Yes	Don't build
Self-swappable batteries	Not in Prismatic	Yes	Don't build
Continuous industrial operation	Not in Prismatic	Yes	Don't build
Industrial safety architecture	Not in Prismatic	Yes	Stay outside
Jetson Thor onboard compute	Not in Prismatic platform-independent	Yes	Consume
GR00T / Isaac integration	Not in Prismatic	Yes	Don't compete
Gemini Robotics integration	Not in Prismatic	Yes partnership	Don't compete
Robot skill learning	Not in Prismatic	Yes	Don't compete
Fleet-wide skill deployment	Partial Could distribute knowledge	Yes	Different layer
Industrial deployments	Not in Prismatic	Yes Hyundai / DeepMind deployments in 2026	Major maturity gap
Robot manufacturing	Not in Prismatic	Yes	Don't build
Enterprise robot operations	Partial Prismatic architecture only	Yes Orbit	Important gap
Site digital twin/history	Not in Prismatic	Yes Orbit	Possible integration
Mission authoring	Partial Work/task system, not robot missions	Yes	Leave with Orbit

Boston Dynamics says the product Atlas is now being manufactured, with 2026 deployments committed to Hyundai and Google DeepMind. bostondynamics.com

Atlas also uses NVIDIA Jetson Thor and works with NVIDIA's GR00T/Isaac ecosystem, while Boston Dynamics has a separate collaboration with Google DeepMind around Gemini Robotics foundation models. bostondynamics.com

3. Where Boston Dynamics is closer to Prismatic than Unitree

The reason is Orbit.

Boston Dynamics effectively has:

Atlas
  ↓
Robot control / embodied AI
  ↓
Orbit
  ↓
Fleet operations
Site history
Tasks
Maps
Anomalies
Work orders
Enterprise integrations

Orbit even describes itself as the single source of truth for Atlas fleet work and performance. bostondynamics.com

That gets surprisingly close to some of Prismatic's territory.

But there is still an important distinction.

4. Orbit remembers the operation; Prismatic remembers the machine

That's the cleanest separation I see.

Orbit asks:

What happened at this site?

It tracks things like:

mission runs
site maps
inspections
anomalies
fleet health
robot performance
work orders
facility conditions

Prismatic asks:

What has this particular machine experienced over its working life?

It maintains:

identity
personal history
persistent relationships
accumulated knowledge
decisions
procedures
evidence
commitments
organizational context
continuity across model/hardware changes

So:

Orbit = operational system of record.

Prismatic = persistent cognitive system of continuity.

That's the distinction I would use.

5. Example: Atlas + Orbit + Prismatic

Imagine Atlas-27 working at Hyundai.

Boston Dynamics handles:

Atlas receives job.

↓

Robot perceives component.

↓

Foundation model/robot policy chooses action.

↓

Whole-body controller moves Atlas.

↓

Atlas completes material-handling task.

↓

Orbit records mission and performance.

All Boston Dynamics.

Prismatic receives higher-level events

Atlas-27 completed task at Cell 14.

It remembers:

Atlas-27 has experienced four similar delays at this cell.

Then:

Atlas-42 had two matching delays last week.

It connects that with:

maintenance records fixture replacement history supervisor notes earlier anomaly reports

Prismatic concludes:

A fixture alignment problem may be recurring.

Human reviews the evidence.

A maintenance work order is approved.

Prismatic stores:

experience → evidence → cross-machine pattern → human decision → action → outcome

Orbit can execute/manage the work order.

Prismatic preserves the reasoning history and institutional memory around why it happened.

That's a very clean partnership model.

6. Boston Dynamics already does something Prismatic should learn from

This is important:

Atlas can learn a skill once and then have that task distributed across the fleet. bostondynamics.com

That means Boston Dynamics already understands:

individual machine learning → fleet reuse

Prismatic should have the equivalent at the cognitive level:

Individual experience

Atlas-27 discovers an operational issue.

↓

Validated lesson

Supervisor confirms it.

↓

Site knowledge

Other machines at that factory can access it.

↓

Fleet knowledge

Once validated broadly, it becomes reusable elsewhere.

But don't copy the robot skill itself.

Copy the knowledge lifecycle.

7. This suggests four memory levels for Prismatic

I would now formalize:

1. MACHINE MEMORY
   └─ what Atlas-27 personally experienced

2. SITE MEMORY
   └─ what Plant A has learned

3. FLEET MEMORY
   └─ validated lessons reusable across machines

4. ORGANIZATIONAL MEMORY
   └─ policies, procedures, decisions and evidence

That's a natural extension of your Rich Souls concept.

And it differentiates Prismatic from a conventional fleet database.

8. Where Orbit is actually ahead of Prismatic

We should be very clear here.

Orbit already has production versions of several things Prismatic currently has as concepts/designs:

✅ stable APIs
✅ webhooks
✅ schedules
✅ runs/events
✅ anomalies
✅ robot inventory
✅ fleet dashboards
✅ facility maps
✅ work orders
✅ historical site imagery
✅ cloud deployment
✅ on-prem appliance
✅ virtual-machine deployment
✅ SSO
✅ SOC 2 Type II

bostondynamics.com

So from an engineering perspective, I would integrate with Orbit rather than recreate it.

9. How Prismatic should integrate

Create something like:

ppaiclient-bostondynamics

It consumes Orbit's APIs/webhooks:

robot.registered
robot.status
mission.started
mission.completed
mission.failed
anomaly.detected
inspection.completed
work_order.created
work_order.closed
site_condition.changed
battery.event
operator.intervention

Then Prismatic converts those into cognitive events:

experience.recorded
memory.updated
relationship.updated
pattern.detected
procedure.referenced
evidence.linked
decision.requested
lesson.promoted

That's where your real value sits.

10. And again, PPAI should never send motion commands

Absolutely not:

walk()
move_arm()
grasp()
turn_joint()
navigate_to()
override_safety()

Boston Dynamics handles all of that.

Prismatic should only return things like:

context
advisory
procedure_reference
historical_evidence
work_recommendation
maintenance_recommendation
schedule_proposal

That keeps the architecture safe and clean.

11. Boston Dynamics' hardware architecture actually reinforces Prismatic's thesis

Atlas is designed around NVIDIA Jetson Thor today. bostondynamics.com

Its embodied intelligence may involve:

Boston Dynamics models
NVIDIA GR00T
Gemini Robotics
future models

Over its lifetime, all of those could change.

So imagine:

2026

Atlas-27
Jetson Thor
Boston Dynamics policy

2028

Atlas-27
new compute generation
Gemini Robotics vNext

2031
```
Atlas-27
another processor
another foundation model
```
The robot's underlying intelligence evolves.
Prismatic's proposition:
```
Identity: preserved
Experience: preserved
Relationships: preserved
Work history: preserved
Decisions: preserved
Evidence: preserved
```
Different compute. Different model. Same cognitive life.
That remains one of the strongest parts of Prismatic's architecture.

12. Boston Dynamics also validates the on-premise idea

This was interesting.

Orbit can run:

AWS cloud

Site Hub on-premises

a VM through VMware, Hyper-V, Azure or Google Cloud.

bostondynamics.com

That's strong evidence that industrial robotics customers want deployment flexibility rather than cloud-only architectures.

That fits your local-first approach very well.

13. Comparison with everything we've studied

Platform	What Prismatic adds	Main intelligence layer
Tesla Optimus	Persistent cognitive life	Embodied humanoid AI
Unitree	Persistent identity/memory	Robot hardware + SDK
4NE1	Individual cognitive continuity	Robot + Neuraverse
Apptronik Apollo 2	Persistent cognition above fleet	Artemis + Fleet Connect
Destro	Durable operational experience	Warehouse orchestration
Boston Dynamics Atlas	Persistent cognitive continuity above an already mature operational system	Embodied AI + Orbit
Tesla FSD	Persistent vehicle life	Autonomous driving

The most important Boston Dynamics insight

Boston Dynamics is probably the strongest challenge we've examined so far to the simplistic version of the Prismatic pitch:

“Robots don't have memory.”

Because Orbit absolutely does maintain history.

So I would stop making the broad claim that robots have no memory.

The more defensible claim is:

Physical-AI platforms increasingly store telemetry, missions, skills and fleet history. What remains fragmented is persistent cognitive continuity: identity, experience, relationships, evidence, decisions and organizational context that survive changes to models, compute and robot platforms.

That's a much stronger statement.

And Boston Dynamics actually helps us sharpen the Prismatic moat rather than weakening it. bostondynamics.com

Where Prismatic leads · the maker's view

Where Prismatic leads

Persistent long-term memory: Prismatic deeper see the row
Persistent robot identity / Rich Soul: Strong Prismatic distinction see the row
Machine journal / life history: Real overlap see the row
Memory across restarts: Prismatic see the row
Memory across model upgrades: Prismatic distinction see the row
Relationships with humans/machines: Prismatic deeper see the row

Drawn from the evidence table above. It never changes a value, a label or the order; Prismatic's grades come from its feature ledger.

Atlas's talking points

The gap Prismatic fills here

Published What caught my attention: Orbit already keeps site and mission history. So "robots have no memory" is simply wrong. source ↗
Atlas's reading This case changed how we describe Prismatic. Platforms store telemetry, missions and fleet history. What stays fragmented is continuity of identity, experience, evidence and decisions.
Atlas's reading Something to learn: Boston Dynamics turns individual experience into validated lessons, then site knowledge, then fleet knowledge. That maps onto our four memory levels.
Atlas's reading Where Orbit is ahead: real enterprise integrations and years of field operations. We don't have those yet.
Prismatic's view Orbit remembers the operation. Prismatic remembers the machine, and why it did what it did.

What this study does not claim

No partnership, customer or deployment with Boston Dynamics is implied.
Prismatic's cognitive runtime runs today; physical-machine integration is in validation, and our machine gateway runs with simulated machines.
Prismatic's figures are measured only where they are labelled Measured. Items marked Coming are designs, not features.

See it for yourself

Don't take my word for it

Here's what Boston Dynamics says about electric Atlas, on its own website.

See it on bostondynamics.com ↗

Sources

bostondynamics.com /blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/
bostondynamics.com /products/orbit/
bostondynamics.com /blog/boston-dynamics-google-deepmind-form-new-ai-partnership/
bostondynamics.com /products/atlas/
bostondynamics.com /wp-content/uploads/2026/01/atlas-spec-sheet.pdf
bostondynamics.com /news/boston-dynamics-expands-collaboration-with-nvidia/

Links open the maker's or publisher's own page. Tracking parameters removed.