Accelerating robotics development with Dragonwing IQ8 and Dragonwing IQ9

Robotics teams building AMRs, mobile manipulators, inspection robots, cobots, or multi-camera perception systems need more than raw TOPS.

They need deterministic control paths, low-latency sensor fusion, industrial I/O, and an edge software stack that can move from prototype to deployment without redesigning the compute layer.

Qualcomm Dragonwing IQ8 and IQ9 are designed for exactly that kind of workload: heterogeneous compute for perception and inference, real-time subsystems for control, and a software stack that supports production-oriented Linux flows and robotics development workflows.

Robotics needs heterogeneous workload-aware compute

Modern robots rarely run as a single monolithic application. Instead, they distribute work across perception pipelines, motion control, safety supervision, and communications, which is why a platform with CPU, GPU, NPU, DSP, and real-time subsystems is so useful.

This architecture accommodates diverse robotics workloads by separating compute-intensive perception and sensor processing from real-time planning and control. With runtimes such as ROS 2, workloads can be allocated to the execution domain best suited to their performance, latency, and determinism requirements.

Dragonwing IQ8 for agile robotics prototypes

Dragonwing IQ8, with its highly efficient 40 dense TOPS architecture, delivers an optimal balance of AI performance, power efficiency, and cost for bringing intelligence to commercial and industrial robots, particularly AMRs.

It enables advanced vision AI, multi-sensor fusion, and industrial connectivity within approachable power and thermal envelopes, helping manufacturers deploy intelligent robotics solutions at scale without the complexity, cost, or power requirements of higher-tier compute platforms.


Dragonwing IQ8 is well suited to commercial and industrial robotics deployments where power, thermal constraints, reliability, and cost matter just as much as AI performance.

For robotics developers, that means Dragonwing IQ8 is a practical base for:

• AMRs with 360 vision using cameras and lidars
  
  
• Indoor inspection robots
  
  
• Cobots with vision-guided pick-and-place.
  
  
• Edge controllers that run local perception and decision-making close to the machine.

A useful pattern is to put frame acquisition, detection, tracking, and segmentation on the AI side, while a companion real-time controller handles motor control, encoder feedback, and safety interlocks.

That reduces latency and keeps time-sensitive control off the general-purpose inference path.

Dragonwing IQ9 for high-throughput autonomy

For robots that need higher AI throughput, more advanced perception, and the compute capacity to run multiple demanding workloads simultaneously, Dragonwing IQ9 is the ideal choice.

IQ9 series deliver up to 100 dense TOPS, with support for demanding industrial deployments and robotics workloads, plus broad I/O expansion for multi-camera systems and production-scale designs.

In robotics terms, Dragonwing IQ9 is the right choice for:

• AMRs with multiple cameras, lidars, and other depth, TOF sensors
  
  
• Outdoor robots with multi-sensor fusion.
  
  
• Warehouse robots combining perception, mapping, and fleet telemetry.
  
  
• Manipulators running multiple concurrent models, such as detection plus pose estimation plus grasp scoring.
  
  
• VLAs

The Dragonwing IQ9 EVK is especially relevant if you want to validate a perception stack under realistic sensor load.

Check the documentation to learn more about support for up to 16 concurrent camera feeds on the IQ-9075 EVK, Robotics SDK coverage, and Gazebo-based robotics simulation with the qrb_ros_simulation package, available as part of the QRB ROS Samples repository for the Dragonwing IQ-9075 Evaluation Kit.

Software stack for robotics

The Dragonwing documentation portal supports two primary development paths: Ubuntu for rapid prototyping, AI/ML development, and early bring-up, and Qualcomm Linux (Yocto) for production-grade embedded deployments with long-term lifecycle requirements.

For robotics teams, that split is valuable:

 - Use Ubuntu for fast iteration on perception, benchmarking, and robot application bring-up.

- Use Qualcomm Linux,  when you need a controlled, lightweight production image with tighter lifecycle management.

Check out the differences between Ubuntu and Qualcomm Linux.

This matters because robotics programs often start in simulation, then move to hardware-in-the-loop testing, then finally to field trials.

A consistent stack across those phases can reduce the “works in sim, breaks on hardware” gap that slows down many autonomy projects.

Building the perception stack

For vision-centric robots, the most important bottleneck is often not the model itself but the pipeline around it: capture, synchronization, pre-processing, inference, post-processing, and routing results to downstream nodes.

In many robotics projects, the real challenge is not training the model but getting cameras, sensors, and ROS nodes working together reliably. Support for these workflows can significantly shorten development and integration cycles.

A typical robotics implementation on Dragonwing starts with its advanced ISP, which enables high-quality multi-camera perception and supports a wide range of camera configurations commonly used in robotics, including stereo, surround-view, and AI vision systems.

From there, the Dragonwing platform intelligently maps perception, AI inference, sensor fusion, tracking, and control workloads to the most appropriate processing resources. Developers do not need to manually orchestrate these engines.

Through familiar robotics frameworks and software tools, the system presents a unified development experience while optimizing performance, power efficiency, and real-time responsiveness behind the scenes.

This allows roboticists to focus on building application capabilities rather than managing underlying compute resources.

From prototype to deployment

One of the most useful aspects of the Dragonwing ecosystem is that Qualcomm is clearly positioning Dragonwing IQ8 and IQ9 as part of a scalable development path rather than isolated boards.

The documentation portal presents the EVKs alongside Qualcomm Linux, Ubuntu, SDKs, and robotics workflows, which supports a cleaner path from lab prototype to productized robot.

For robotics product teams, that means less rework when you move from a dev kit to a custom carrier or module-based design. It also means your autonomy stack can stay focused on behavior, navigation, manipulation, and safety rather than spending excess effort on low-level platform porting.

Where each fits

Platform	Best fit	Robotics strengths
Dragonwing IQ8	Fast-moving robotics prototypes and commercial robots with strong edge AI needs	Vision inference, sensor fusion, moderate multi-camera workloads, Ubuntu/Qualcomm Linux development
Dragonwing IQ9	Higher-end autonomous robots and production-scale deployments	Higher TOPS, more  camera headroom, stronger I/O expansion, heavier multi-model autonomy pipelines.

What’s next

If your robotics roadmap is moving toward real-time perception, multi-sensor fusion, and autonomous behavior at the edge, Dragonwing IQ8 and IQ9 give you a solid hardware and software base to build on.

Dragonwing IQ8 is the pragmatic entry point, while Dragonwing IQ9 gives you the extra performance envelope for more demanding autonomy stacks and multi-camera robots.

Check out our development hardware for Dragonwing IQ9 and Dragonwing IQ8.

Explore our open documentation portal to get started today