Investigation of Autonomous Robotic Path Planning Utilising Automated CAD Generation
TWI Industrial Member Report 1218-2026
By Mark Sutcliffe, Myra Zhang and Sam Hurrell
Industrial Need
While robotic inspection technologies are well established for repetitive and structured tasks, their wider adoption remains constrained by a strong dependence on accurate and up-to-date computer-aided design (CAD) models. In practice, such models are frequently unavailable, outdated, or unreliable due to component repairs, in-service modifications, manufacturing tolerances, or legacy assets where design data no longer exists. This disconnect between digital design intent and physical reality represents a significant barrier to efficient robotic deployment.
In the absence of reliable CAD data, inspection planning becomes a manual, iterative process requiring highly skilled robotic and NDT specialists. Operators must assess component geometry, identify deviations, and repeatedly adjust inspection paths to ensure adequate coverage and probe access. For complex or safety-critical components, this process can take many hours or even days, leading to increased inspection costs. These challenges are particularly acute where inspection volumes are high and geometries are often complex or bespoke.
There is a clear industry need for robotic inspection systems that can adapt to unknown geometries with minimal human intervention. Such systems must be capable of rapidly digitising physical components, autonomously generating inspection paths, and aligning planned trajectories to real-world assets. Reducing dependence on specialist knowledge not only improves efficiency and consistency but also addresses skills shortages that are increasingly affecting engineering and inspection disciplines.
Addressing these needs requires the integration of advanced sensing technologies, intelligent path-planning algorithms, and Artificial Intelligence (AI) assisted decision-making. Solutions that can shorten planning times from days to minutes, while maintaining inspection quality and coverage, offer significant commercial and operational value. The development of such adaptive, sensor-driven robotic inspection capabilities is therefore essential to improving productivity, safety, and resilience across industries.
Key Findings
- Robotic inspection achieved without pre-existing CAD models. The research demonstrated a complete sensor-driven workflow capable of generating CAD models, planning inspection paths, and executing robotic scans with minimal operator input, significantly reducing dependence on accurate CAD data.
- Optical 3-Dimensional (3D) scanning provided fast and reliable CAD generation. Compared with low-cost Light Detection And Ranging (LiDAR) sensor, the ZG optical scanning system delivered superior robustness, rapid point-cloud stitching, and reliable point-cloud-to-mesh conversion, enabling CAD generation within minutes under varying lighting conditions.
- A novel graph-based algorithmic path-planning approach was developed and demonstrated as effective and practical. The hybrid graph-theory method developed within this project guaranteed full surface coverage and supported obstacle avoidance making it suitable for real-world deployment and industrial use.
- Fully AI-driven path planning method developed, implemented and validated. The Neuro Evolution of Augmenting Topologies (NEAT) based learning approach successfully adapted to complex 2-Dimensional (2D) and 3D geometries in simulation but required long training times, preventing its use in live robotic trials within this project.
- A sensor driven path alignment method was developed. A three-point calibration method using laser depth sensing and visual feedback enabled reliable scaling, rotation, and alignment of planned paths to physical components, supporting remote and mobile inspection operations.
Impact
This research delivers a significant advancement in robotic inspection by enabling automated inspection in situations where accurate CAD models are unavailable or incomplete. By integrating advanced 3D sensing, algorithmic path planning, and AI-based learning methods, the project reduces reliance on highly skilled operators and compresses inspection-planning times from hours or days to minutes. This has direct implications for cost reduction, improved productivity, and increased deployment flexibility in manufacturing and maintenance environments.
The successful demonstration of an end-to-end, sensor-driven inspection workflow shows that robotic inspection systems can rapidly adapt to unknown or modified geometries, supporting real-world scenarios such as repaired components, legacy assets, and in-service structures. The novel graph-based path-planning algorithm provides guaranteed surface coverage, computational efficiency, and ease of deployment, making it immediately applicable to industrial use.
The inclusion of a remote alignment and correction system enhances operational safety by enabling inspections to be conducted at a distance, reducing human exposure to hazardous environments. This is particularly relevant for large or complex components where manual access is limited or unsafe.
AI-driven path planning demonstrates strong potential for future adoption as computing resources improve. Overall, this work strengthens TWI’s technical capability in autonomous robotic inspection and establishes a scalable foundation for future developments in intelligent, mobile systems, supporting increased efficiency, safety, and adaptability across multiple industrial sectors.