An Automated Way to Assemble Complex Real-World Objects
Published on : Thursday 07-09-2023
Planning for mechanical assemblies necessitates more than simply sketching up rough drawings, says Nivesh.
The manufacturing industry has widely embraced AI. Planning for mechanical assemblies necessitates more than simply sketching up rough drawings. It is a multifaceted challenge that involves working with intricate 3D shapes and addressing the stringent motion constraints inherent in real-world assemblies.
It is comprehensible that human engineers are required to actively engage in the process of designing assembly plans and instructions before transmitting the components to assembly lines. However, this manual approach is associated with elevated labour expenses and possible mistakes.
Dataset
The researchers generated a comprehensive dataset consisting of numerous industrially relevant assemblies and motions designed to evaluate the efficacy of their proposed strategy. The method under consideration demonstrates a high level of effectiveness in addressing a wide range of challenges, mainly exhibiting superior performance compared to prior methodologies, particularly in the context of rotational assemblies such as screws and puzzles. Moreover, this software has remarkable computational efficiency as it can solve complex 80-part assemblies in minutes.
Disassembly planning
The method aims to determine the optimal assembly strategy for attaching a screw to a rod. This process involves two distinct stages: disassembly and assembly. The technique for disassembly planning aims to identify a path free from any potential collisions, hence facilitating the removal of the screw from the rod. The algorithm applies diverse forces to the screw by employing physics-based simulation and subsequently observes its resultant movement.
Thus, supplying torque at the rod's central axis causes rotational motion, which moves the screw down the rod. Providing a linear force outward from the rod separates the screw from the rod. During the assembly stage, the algorithm reverses the disassembly path to obtain an assembly solution by reassembling the individual parts.
The process of disassembling and assembling
In contemporary manufacturing processes, particularly in factory or assembly line settings, it is customary for all operational instructions to be pre-programmed fixedly. To successfully construct a given product, it is essential to exercise exact control over the instructions for the assembly and disassembly of said object. Which component should be assembled initially? Which component should be assembled subsequently? Furthermore, what is your plan for the process of assembling this?
Previous endeavours have predominantly focused on elementary assembly pathways, characterised by a straightforward conversion of components, without delving into intricate complexities. The team employed a physics-based simulator to progress beyond the current state, a widely utilised tool for training autonomous systems such as robotics and self-driving automobiles. This simulator facilitated the exploration of assembly paths, enhancing the process's ease and generalisability.
Implementation
Implementing their technique, which involves disassembling and reassembling, restricts the assemblies to a state of rigidity exclusively. Nevertheless, geometric-based methodologies need to be improved in their ability to effectively handle deformable objects due to their inability to represent the physical deformations accurately. Hence, the researchers find it intriguing to delve deeper into physics-based planning to investigate its potential for generalisation in deformable assemblies, such as the snap-fit assembly.
Furthermore, utilising geometric data and tactile feedback is imperative while devising assembly plans. In contrast to the exhaustive search strategy employed by their technique and baseline methods, humans can rapidly deduce probable disassembly sequences and motions based on visual cues. It enables them to avoid spending unnecessary effort attempting to disassemble obstructed pieces or moving parts in directions that lead to dead ends.
Conclusion
To enhance research efforts in robotic assembly, a promising avenue for future exploration involves incorporating robotic arms into simulation environments.
This integration would enable the manipulation of assemblies by the predetermined paths generated by the proposed approach. The research team posits that there is promise in expanding the capacity for autonomous and flexible execution of intricate assembly tasks on tangible robotic systems.
Furthermore, the researchers are considering creating a physical robotic system to assemble diverse things. The successful integration of automatic control and planning into the team's system necessitates additional effort, serving as a crucial step towards their overarching objective of developing a fully autonomous assembly line capable of adaptively assembling a wide range of products without human intervention.
References
1. https://arxiv.org/abs/2211.03977
2. https://www.industrialautomationindia.in/articleitm/13628/Building-a-Smart-Factory-I:-From-Automation-to-Autonomy%C2%A0%C2%A0-%C2%A0/articles
3. https://news.mit.edu/2022/automated-way-assemble-thousands-objects-1207
