Doosan Robotics partners with Daedong to build on-device AI for agricultural robots
Doosan Robotics and agritech manufacturer Daedong have signed an MOU to co-develop field robots for smart farms and outdoor agricultural work. The partnership centers on on-device AI, enabling robots to run perception and control locally with low latency and no dependency on cloud connectivity.
What's being built
- Mobile Manipulator (MoMa) for agriculture: Doosan will design and manufacture the robotic arms, control systems, and motion planning for manipulation. Daedong will deliver the autonomous mobility platform, leveraging its field equipment expertise.
- On-device AI for unstructured environments: The teams will develop AI that can recognize at a human level and execute tasks despite dust, glare, occlusions, uneven ground, and changing crop conditions.
- Commercialization and market expansion: The collaboration targets global rollout of field robots, with Daedong focusing on field tests, optimization, safety certifications, and support for overseas regulatory compliance.
Why on-device AI matters in agriculture
Fields don't guarantee bandwidth or stable connectivity. Running inference on-device keeps perception and control responsive, even offline. It also reduces data exposure by keeping raw video and sensor data local, while still enabling over-the-air updates when connectivity is available.
For developers, this implies tight performance budgets, model optimization, and robust fallback strategies. Expect aggressive use of quantization, pruning, and hardware acceleration to maintain real-time operation on power-limited platforms.
Division of responsibilities
- Doosan Robotics: Robotic arm design and manufacturing, arm control systems, motion development, and standardization of SoC application performance, functionality, and interfaces. Development of AI algorithms and firmware for on-device operation.
- Daedong: Autonomous mobility platform design and manufacturing built on diverse agricultural field data. Lead field trials and tuning, safety certification for ag environments, and support for overseas compliance.
Technical focus areas to watch
- SoC standardization: Consistent interfaces for sensors, actuators, and accelerators (GPU/NPU/DSP), plus well-defined performance targets per task (e.g., perception, planning, grasping).
- Perception in unstructured scenes: Multi-sensor fusion (RGB, depth, LiDAR), lighting robustness, occlusion handling, and domain adaptation across crop types and seasons.
- Motion + mobility: Coordinated control between base and arm for manipulation while moving (MoMa), safe human-robot interaction, and terrain-aware navigation.
- Safety + reliability: Redundant sensing, E-stop paths, degraded modes when sensors fail, and validation datasets sourced from real fields.
- Tooling and deployment: Model conversion/compilation pipelines, deterministic scheduling, OTA updates that tolerate spotty connectivity, and telemetry with privacy-aware logging.
For IT and development teams
- Expect SDKs/APIs exposing arm kinematics, mobility, and perception, likely with ROS 2 integration for faster prototyping and orchestration. See ROS 2 docs.
- Plan for edge inference with ONNX export and hardware-targeted runtimes. A common path is ONNX + accelerator-specific compilers; see ONNX Runtime.
- Data pipeline: capture-label-train-optimize cycles from real fields. Include night, dust, rain, mud, and seasonal variation.
- Ops: rollout strategies for fleets, staged updates, remote diagnostics, and safe rollback.
Government-backed SoC projects
The companies plan to pursue government-led AI SoC initiatives. Expect emphasis on energy-efficient inference, standardized interfaces, and long-term support lifecycles suitable for industrial deployments.
Global commercialization and compliance
Daedong will drive field testing, optimization, and certifications tailored to agriculture. That includes safety in proximity to people, crops, and livestock, plus documentation to meet overseas regulatory requirements.
Leadership context
Following the MOU signing, Kevin Kim, CEO of Doosan Robotics, emphasized solving labor shortages and harsh field conditions through faster commercialization of MoMa with Daedong's equipment expertise. Yoohyun Won, vice chairman of Daedong, joined him for the commemorative photo at the signing.
Bottom line
This partnership blends industrial-grade manipulation from Doosan with Daedong's mobility and field know-how. For developers, it points to a practical, on-device AI stack for real farms-where latency, uptime, and safety matter more than lab benchmarks.
If you're building skills in edge AI and robotics, explore curated learning paths at Complete AI Training.
Your membership also unlocks:
