Traditional industrial robots are highly effective in structured and repetitive environments, but they often require fixed fixtures, predefined CAD models, manually programmed trajectories and consistent part positioning. These requirements limit their flexibility in high-mix, low-volume production, rework, repair and other applications where every workpiece may be slightly different.
Scan&Go 2.0, developed by Maple Advanced Robotics Inc. (MARI), addresses this challenge through adaptive path planning, which enables robots to understand changing workpiece conditions by automatically generating task-specific trajectories from real-world visual information.
Instead of asking an engineer to manually program every path, Scan&Go 2.0 allows an operator to directly indicate what needs to be processed. The system then interprets the operator’s intent, analyzes the surface and generates the corresponding robot path.
Programming and planning
Conventional offline programming typically begins with a CAD model and assumes that the actual part closely matches the digital design. In real production, however, parts may shift, deform, vary in dimension, contain localized defects or require different processing areas.
Scan&Go 2.0 generates robot trajectories directly from 2-D and 3-D sensor data. By combining machine vision, point-cloud processing, process knowledge and physics-based planning, the system can adapt the robot path to the actual condition of each workpiece.
This creates a more flexible workflow: Scan the part. Identify the target area. Generate the path. Execute the process.
The robot is no longer limited to repeating a previously programmed motion. It can dynamically plan how to perform the task based on what it sees.
Spot sanding
In automotive repair, surface finishing and quality correction, workers often mark defects directly on the part using a pen, sticker, tape or another visual indicator.
Scan&Go 2.0 can recognize these marked regions and automatically generate localized spot sanding trajectories around them. Instead of manually teaching the robot each defect location, the operator can simply mark the areas requiring correction.
The system identifies the markers, maps them onto the 3-D surface, defines the appropriate region of interest and creates a sanding path that follows the local surface geometry. This makes human-robot interaction much more direct. The operator communicates the task through a familiar physical action while the robot translates that instruction into an accurate and executable trajectory.
Bump grinding
Many manufactured parts contain localized raised features, including weld beads, excess adhesive, casting flash, surface bumps and other unwanted protrusions. Scan&Go 2.0 can analyze 3-D surface data to identify these elevated regions and generate targeted bump grinding paths. Rather than grinding the entire surface, the robot focuses only on the areas that exceed the desired surface profile.
The adaptive path can follow the shape, boundary, orientation and height variation of the detected bump. This selective approach can reduce cycle time, abrasive consumption, unnecessary material removal and the risk of damaging acceptable surrounding surfaces.
Because the trajectory is generated from the measured condition of the actual part, the system can also respond to variations between workpieces without requiring a new robot program for every part.
Dynamic local paths
Scan&Go 2.0 also supports adaptive processing based on differences in color, texture, coating or visible surface appearance.
For example, the system may identify:
- A painted region requiring polishing
- A different-colored patch requiring removal
- A repair primer area requiring sanding
- Excess adhesive with a distinct visual appearance
- A coated or contaminated section requiring localized treatment
Once the target color region is detected, the system projects the 2-D visual information onto the 3-D surface model. It then dynamically creates a local path that follows the detected boundary and actual surface curvature.
This enables operators to define work regions through visible surface characteristics instead of relying only on CAD geometry or fixed coordinates.
Curved-surface welding
Scan&Go 2.0 also extends adaptive path planning to complex curved-surface welding applications. By analyzing the 3-D geometry of intersecting surfaces, the system can automatically identify the joint location and accurately generate a welding path along the actual intersection line.
This eliminates the need to rely solely on nominal CAD geometry or manually taught robot points.
This capability is especially important when welding large or flexible components where heat, forming processes, assembly tolerances or material deformation may cause the actual joint to deviate significantly from its original design. Even when deformation creates uneven gaps, Scan&Go 2.0 can use measured surface data to adapt the welding trajectory, tool orientation and process parameters to the real condition of the workpiece.
Beyond single-pass welding, Scan&Go 2.0 supports true multi-layer, multi-pass welding. Based on the detected joint geometry and required weld volume, the system can plan multiple coordinated welding passes, define the position and sequence of each layer, and adjust subsequent trajectories according to the evolving weld profile. This creates a foundation for automating demanding applications such as pressure vessels, pipelines and other large, curved structures where joint geometry and material conditions vary from part to part.
Physics AI
The key value of the system is not simply visual detection. Its larger advantage lies in how it converts human intent and real-world surface information into a physically executable robot process.
Scan&Go 2.0 uses Physics AI to incorporate process constraints into path planning, including surface orientation, tool angle, contact force, robot reachability, motion continuity, tool dimensions, edge conditions and process-specific requirements.
The system, therefore, does more than identify where the robot should work. It also determines how the robot should approach, contact, move across and leave the target surface.
This creates a more natural form of human-robot interaction. An operator can indicate the required task using a marker, a visual feature, a selected region or the physical condition of the part. The system then converts that instruction into a robot-ready process.
The operator focuses on the manufacturing objective, while the system handles the complex steps of perception, coordinate transformation, path generation and motion execution.
Force-aware execution
Adaptive path planning delivers value only when the generated trajectory can be translated into accurate and consistent physical motion. Doosan Robotics provides the execution platform that allows Scan&Go 2.0 to turn visual understanding and process intelligence into reliable real-world operations.
Scan&Go 2.0 uses proprietary 3-D vision and physics-informed AI to model the actual workpiece and generate task-specific paths with up to 0.5-mm planning precision. The Doosan collaborative robot then translates these computational paths into controlled physical motion through advanced force and compliance control, enabled by torque sensors integrated into every joint.
This joint-level sensing allows the robot to continuously detect and respond to variations in surface height, curvature, orientation and contact conditions.
Rather than rigidly following a fixed trajectory, the robot can compliantly adjust its motion and maintain the required tool contact as the actual surface changes. This is particularly important in applications such as sanding, grinding, polishing and surface finishing where tool orientation, contact force and motion continuity directly affect process quality.
Safety is integrated at the system level rather than treated as a separate consideration. Scan&Go 2.0 is designed to meet PL e, Category 4 requirements and incorporates safety-rated stop functions and collision detection, supporting responsive operation in human-robot collaborative environments.
Together, Scan&Go 2.0 and Doosan Robotics create a complete adaptive automation workflow: Scan&Go 2.0 understands the workpiece and determines how the task should be performed while the Doosan robot carries out the process accurately, consistently and safely.
Human-robot interface
In traditional automation, communication between a human and a robot is indirect. The operator explains the requirement to a robot programmer, the programmer creates points and trajectories, and the robot executes the resulting code. Scan&Go 2.0, however, shortens this interaction chain.
A worker can mark a defect. The vision system identifies it. The system generates the local path. The robot performs the correction.
This approach allows process experts to guide robots without becoming robot programming experts. It also helps preserve the value of human judgment while reducing the repetitive, dusty and physically demanding work associated with sanding, grinding, polishing and repair.
The human determines what needs to be done. The robot determines how to execute it consistently.
Enabling AI
Scan&Go 2.0 represents a practical step forward in adaptive Physical AI manufacturing. Developed by MARI in collaboration with Doosan Robotics, the technology was recognized with the CES Best of Innovation Award in Artificial Intelligence at the 2026 Consumer Electronics Show in Las Vegas, demonstrating its potential to bring intelligent perception, adaptive planning and autonomous execution into real industrial environments.
Its adaptive path planning capabilities are particularly valuable in applications where part geometry, defect location and processing requirements vary from one workpiece to another. Representative applications include the repair and finishing of wind turbine blades, localized surface correction on automotive body panels and precision coating sanding on aircraft radomes. These tasks demand far more adaptability than conventional fixed-path automation because every surface condition, repair area and required trajectory may be different.
By combining visual understanding, 3-D surface analysis, Physics AI, force-controlled execution and dynamic local path generation, Scan&Go 2.0 allows robots to respond directly to actual workpiece conditions and human instructions. It transforms robot programming from a specialized engineering activity into a more intuitive interaction between operators, processes and intelligent machines – making advanced automation more accessible, flexible and effective in some of manufacturing’s most demanding applications.
Read more about this topic in our extensive automation & robotics archive.
