edge0

Robot Chips That Ship
With On-Device Intelligence

Chip + Model Co-Delivery≤220ms Command Response150MB Runtime FootprintOffline-First Operation

D-Robotics serves embodied AI and small-to-mid-size robots with on-device intelligence across its X and S series chips. edge0 co-tunes on-device audio models, natural-language understanding and the chip runtime, so vendors get local reasoning, low latency and offline-ready interaction through “chip + on-device model”. Robots hold natural conversations, understand context and call local action interfaces — without relying on the cloud.

X & S Series
Completed Locally · ≤220msVoiceCommand ParsingRobot Action
X / S
Chip Series Adapted
From flagship to cost-optimized SKUs
≤220ms
Local Command Response
P95, from end of speech to action callback
150MB
Memory Footprint
Peak runtime memory

Downstream robot makers need edge-side interaction that works out of the box. By co-tuning edge0’s on-device model, runtime and command interface with our X and S series, robot vendors can unplug the network at delivery and still get natural voice expression and timely action execution — and the S series can still call cloud LLMs when a task needs them, which gives us a lower-cost, more natural voice interaction story.

Why On-Device Models?

Robots Work in the Field,
Core Commands Can’t Wait for the Network

In warehouses, yards and public spaces, latency, cost and network reliability are hard constraints at the execution layer. Keeping core commands on-device turns interaction into a product capability instead of a network gamble.

01

Usable the Moment It’s Unboxed

The chip and the on-device model are delivered and validated as one unit, so downstream vendors skip a separate AI integration phase entirely.

Chip + Model Co-Delivery
02

Motion Commands Answer in Milliseconds

Wakeup, recognition and intent parsing all finish on the chip, so a spoken command turns into motion without a cloud round trip.

P95 Response ≤220ms
03

No Network, No Failure Point

In weak-signal or fully offline environments, core commands stay available and the robot keeps completing the task in front of it.

Core Commands 100% Local
All-Cloud Path
RobotNetworkCloudReturn LinkAction
1000–2500ms
X & S Series Local Path
VoiceOn-Device ModelRobot APIAction
≤220ms
Why edge0?

Customers Choose edge0 Because the Model
Runs Directly on X and S Series Chips

On the edge side, edge0 targets the compute and memory envelope of mainstream robot chips. “Out of the box” means usable interaction capability at chip delivery — not after a months-long integration project.

X / S
Chip Series Adapted
≤220ms
Command Response P95
150MB
Memory Footprint
1h
To First On-Device Demo
01

Model and Chip Co-Tuned

Quantization, memory planning and scheduling are tuned against the X and S series NPUs, so the model fits the silicon it ships on.

02

Full Local Voice Pipeline

Wakeup, ASR and intent understanding run end to end on-device, with no cloud dependency in the core loop.

03

Standard Embodied Command Set

A structured action-control interface maps intents to robot motions, so vendors integrate commands instead of raw models.

04

Chip and Model Delivered Together

One joint delivery package covers silicon, model, runtime and documentation, cutting evaluation and integration cycles.

Downstream Vendors Get Three Capabilities:
Real-Time Interaction, Local Control, Offline Operation

Local Real-Time Interaction
Memory Footprint
RAM ≤ 4GB
Local real-time interaction (ASR+LLM+TTS):
end-to-end under 800ms
Local Command Control
≤220ms
User Wakeup
0ms
Chip Processing
ASR → NLU
Start Moving
≤220ms
No cloud API call in the loop
Offline Operation
“Return to the dock.”
Voice
On-device
Chip Processing
Intent → Plan
Robot Action
Executed
LOCAL inference · 0 API calls · READY action complete

Same Voice Interaction,
a New Boundary of What’s Usable

Key Metric
Cloud-Only Approach
Chip + edge0 On-Device Model
Integration Starting Point
Cloud connectivity and accounts come first
X & S chips ship with edge capability built in
Core Commands
Bound to network and cloud availability
Execute locally, offline and on weak links
Command Response
1000–2500ms round trips feel laggy
P95 ≤220ms, local action callback
Cloud Cost
Billed per call, grows with fleet size
High-frequency commands on-device, cloud cost −80%
Downstream Development
Cloud-side logic to build and maintain
Standard SDK and command set simplify integration

Subscribe for the latest news on Edge0 products