ROBOTICS
Designing environments that are robot-inclusive
SINGAPORE UNIVERSITY OF TECHNOLOGY AND DESIGN
IMAGE:
AN OVERVIEW OF THE DIGITAL TWIN SYSTEM PROPOSED FOR EVALUATING ROBOT-INCLUSIVITY IN THE BUILT ENVIRONMENT
view moreCREDIT: SUTD
Humans and robots are increasingly interacting within built environments such as cities, buildings, walkways, and parks. Offering adaptability, cost-effectiveness, and scalability, robots are gradually being integrated into various aspects of everyday life, from manufacturing to healthcare to hospitality.
“Ensuring that robots can navigate and operate effectively within built environments is crucial for their widespread adoption and acceptance,” said Associate Professor Mohan Rajesh Elara from the Singapore University of Technology and Design (SUTD).
To have fully autonomous service robots operate in human environments, however, is still a distant goal. Spatial limitations in the built environment restrict a robot’s performance capability. In designing robot-inclusive environments, robot interaction within a built environment must be examined. The current methods used for this involve real-life testing and physical experiments that are costly, time-consuming, and labour-intensive.
To address these limitations, Assoc Prof Mohan and his SUTD team explored an innovative approach in their paper ‘Enhancing robot inclusivity in the built environment: A digital twin-assisted assessment of design guideline compliance’. Here, they demonstrate a novel methodology utilising digital twins to establish the usefulness of built environment design guidelines for robots. They also model some robot archetypes and environments as digital twins to examine robot behaviour within the environments.
A digital twin is a virtual replica of a physical object in a virtual version of its environment. “The digital twin approach offers several key advantages, including the ability to simulate real-world scenarios, enable virtual testing of robot interactions, and provide insights into compliance with design guidelines before physical implementation,” said Assoc Prof Mohan. Moreover, using digital twins allows real-time monitoring, hazard identification, and training a robot’s algorithm before deployment.
In the study, Assoc Prof Mohan uses digital twins to analyse the robot-friendliness of the built environment and prepare for robot deployment. The methodology used is divided into three phases: documentation, digitisation, and design analysis.
First, on-site documentation of the environment is necessary for the simulation. It can be done via direct data collection, laser scanning, or photogrammetry techniques. Ideally performed during the building’s design process, direct data collection uses Building Information Modelling (BIM)—a process of generating and managing digital representations of the building. When a building has already been constructed, laser scanning or photogrammetry techniques can be used to generate point cloud data for processing.
Second, digitisation focuses on making the built environment’s digital model suitable for the robot simulation software. In this step, point cloud data will be reconstructed into a digital space and used to generate three-dimensional (3D) models of the built environment.
Finally, the digital model is designed and analysed. Using the digitised model of the environment in the robot simulation software, the behaviours and interactions of various robots are tested within the environment. Virtual scenarios are made based on existing design guidelines of built environments, and the robots are assessed on their navigation, path planning, and interaction with the surrounding.
In one case study, Assoc Prof Mohan used digital twins to test four different cleaning robots in six different environments that adhered to Accessibility Design Guidelines. Of the four robots, one completed the most goals and performed the best in the simulated environments. It is important to note that robot inclusiveness does not always translate to robot performance efficiency. However, an inclusive environment does promote better accessibility for robots, allowing them to complete their tasks properly.
With robots increasingly being used in urban applications such as cleaning, logistics, and building maintenance, this study’s findings will help improve design guidelines for built environments to accommodate robots. Better design guidelines will allow the seamless integration of robots into human-centric spaces and their enhanced efficiency in various applications.
“The findings could shape future space design by emphasising flexibility, adaptability, and accessibility to accommodate robot interactions,” Assoc Prof Mohan adds.
In the future, the research team aims to extend the current methods and autonomously generate the infrastructure modifications required to improve the accessibility of mobile robots through the use of design, AI and technology. Assoc Prof Mohan also hopes to develop a set of design guidelines and recommendations for building robot-friendly infrastructure.
JOURNAL
Buildings
ARTICLE TITLE
Enhancing robot inclusivity in the built environment: A digital twin-assisted assessment of design guideline compliance
Navigating new horizons: Pioneering AI
framework enhances robot efficiency
and planning
IMAGE:
SCIENTISTS AT THE INSTITUTE OF FUTURE TECHNOLOGY AT SHANGHAI UNIVERSITY HAVE USED LARGE LANGUAGE MODELS (LLMS) TO IMPROVE THE EFFICIENCY AND EFFECTIVENESS OF ROBOTS PERFORMING COMPLEX, INSTRUCTION-BASED TASKS.
view moreCREDIT: YUAN ZHANG, SCHOOL OF FUTURE TECHNOLOGY, SHANGHAI UNIVERSITY
In a groundbreaking study published in Cyborg Bionic Systems, researchers from Shanghai University have unveiled a new artificial intelligence framework that revolutionizes the way robots interpret and execute tasks. The "Correction and Planning with Memory Integration" (CPMI) framework leverages large language models (LLMs) to improve the efficiency and effectiveness of robots performing complex, instruction-based tasks.
Traditionally, robots required explicit programming and extensive data to navigate and interact with their environment, often struggling with unexpected challenges or changes in their tasks. However, the team, led by Yuan Zhang and Chao Wang, has introduced a dynamic new approach that integrates memory and planning capabilities within LLMs, enabling robots to adapt and learn from their experiences in real-time.
A Leap Forward in Robotic Task Management
The CPMI framework marks a significant departure from conventional methods by using LLMs not just as tools for processing language but as central decision-making elements in robotic tasks. This innovative use of AI allows robots to break down complex instructions into actionable steps, plan their actions more effectively, and correct their course in response to obstacles or errors.
One of the most striking features of the CPMI framework is its memory module, which gives robots the ability to remember and learn from previous tasks. This capability mimics human memory and experience, enabling robots to perform more efficiently over time and adapt to new situations with unprecedented speed.
Demonstrating Superior Performance
The research team tested their framework using the ALFRED simulation environment, where it outperformed existing models in "few-shot" scenarios—situations where robots have limited examples to learn from. The CPMI framework not only achieved higher success rates but also demonstrated significant improvements in task efficiency and adaptability.
"By integrating memory and planning within a single AI-driven framework, we have enabled robots to learn from each interaction and improve their decision-making processes continuously," explained Chao Wang, the corresponding author of the study. "This not only enhances their performance but also reduces the need for extensive pre-programming and data collection."
Future Applications and Developments
The potential applications for the CPMI framework are vast, ranging from domestic robots that can better assist in household tasks to industrial robots that can navigate complex manufacturing processes. As LLMs continue to evolve, the capabilities of CPMI-equipped robots are expected to grow, leading to more autonomous and intelligent machines.
The Shanghai University team is optimistic about the future of robotic technology and plans to continue refining their framework. "Our next steps involve enhancing the memory capabilities of the CPMI framework and testing it in more diverse and challenging environments," said Yuan Zhang. "We believe that this technology has the potential to transform not just robotics but any field that relies on complex, real-time decision-making."
This research not only sets a new standard for AI in robotics but also opens up new pathways for the integration of advanced AI technologies in everyday life. With the continued development of frameworks like CPMI, the dream of having intelligent, adaptable robots that can perform a wide range of tasks effectively and independently is becoming a tangible reality.
The paper, "Leave It to Large Language Models! Correction and Planning with Memory Integration," was published in the journal Cyborg and Bionic Systems on Mar 27,2024, at DOI: https://spj.science.org/doi/10.34133/cbsystems.0087
JOURNAL
Cyborg and Bionic Systems
ARTICLE TITLE
Leave It to Large Language Models! Correction and Planning with Memory Integration
No comments:
Post a Comment