The Robotics Roundup is a weekly newspost going over some of the most exciting developments in robotics over the past week.
In today’s edition we have:
- Improving artificial intelligence with games
- Plastic ‘Muscle’ Pumps Up Soft Robotics
- Household Robot Navigation System Developed by MIT Researchers
- Planetary analog environments explored by a team of legged robots
- GPTs And Robotics: Why It’s Crucial To Create Failsafes
Improving artificial intelligence with games
Researchers are shifting their focus to the challenges posed by modern video games as a way to advance machine intelligence. While computers have surpassed human skill in strategy games like checkers, chess, and Go, video games offer a wider range of skills to practice. Open-world video games present unsolved challenges, requiring AI systems to learn diverse skills and apply them across complex environments. Using video games as research platforms benefits the gaming industry and holds promise for the development of the metaverse. Progress in AI within video games contributes to fundamental AI topics and facilitates the transfer of techniques from games to the real world.
Plastic ‘Muscle’ Pumps Up Soft Robotics
Researchers at Penn State have developed a new ferroelectric polymer that could enable robots to use plastic “muscles”. The ferroelectric polymer nanocomposites outperform traditional piezoelectric polymer composites in creating high-performance motion controllers. The new material allows for soft actuators that can replace rigid materials, providing greater flexibility and mimicking human muscle. The researchers overcame the challenge of high driving fields by creating a percolative ferroelectric polymer nanocomposite, resulting in shape change with significantly less energy compared to traditional ferroelectric materials. This breakthrough could be used in various applications that require low driving fields, such as medical devices, optical devices, and soft robotics.
Household Robot Navigation System Developed by MIT Researchers
MIT researchers have developed a neural network program called PIGINet (Plans, Images, Goal and Initial facts) that brings task and motion planning capabilities to domestic robots. The system uses a learning-enabled Task and Motion Planning (TAMP) algorithm to adapt robot movements to handle movable obstacles in home environments. In experiments, the platform reduced runtime by 80% in small spaces and by 10%-50% in larger areas. While currently focused on kitchen spaces, the researchers aim to expand to other rooms in the house and potentially outside of the domestic space in the future.
Planetary analog environments explored by a team of legged robots
Swiss researchers from ETH Zurich are exploring the idea of sending a team of robots to explore the moon instead of a single rover. The researchers equipped three ANYmal legged robots with various measuring and analysis instruments to make them suitable for lunar exploration. The robots were tested in Switzerland and at the European Space Resources Innovation Center (ESRIC) in Luxembourg, where they won a European competition for lunar exploration robots. The team of robots includes specialists and a generalist, with redundancy built in to compensate for any failures. The researchers also plan to make the robots more autonomous in the future.
GPTs And Robotics: Why It’s Crucial To Create Failsafes
GPT AIs, such as ChatGPT, have shown proficiency in generating and debugging code, making them useful in robotics. By integrating GPT AI with robotics, complex coding tasks that would usually take months or years can be completed in a matter of days or weeks. However, the connection between GPT AI and the physical world via robotics carries risks, especially with larger robotic platforms that have the potential to cause harm. To mitigate these risks, it is important that we develop safeguards and failsafe systems prior to deployments of potentially dangerous systems.