India's first commercial walking robot Stoch uses machine learning to move around
The robot is the size of a small dog with four legs and a thick but flexible spine. When connected to a battery, it starts walking on slender, articulated limbs, like a canine.
It even has a 'face' that looks vaguely like that of a pug. The developers at the Robert Bosch Centre for Cyber-physical Systems at the Indian Institute of Science (IISc) are calling it 'Stoch' and they say it's on target to become the country's first commercial "walking" robot.
Stoch has been under development since January last year. The first version was displayed a month ago at Aero India - it was heavy-footed and clumsy. A sleeker second version was developed just two weeks ago and a third will be ready in three months. A commercial variant is about a year away.
The robot uses machine learning to figure out how to walk by itself. Specifically, it uses reinforcement learning, where the machine learns over time to take the best possible action in return for the best reward.
After several million attempts - performed in computer simulation - ؙthe robot learns to walk. Some Indian Institutes of Technology (IITs) and IISc have been working on such devices in the past few years. There aren't any commercial versions in the market and no institution other than IISc has a prototype that uses reinforcement learning to teach the robot to walk.
The IISc project started when a student, Shounak Bhattacharya, did a master's project in the department of mechanical engineering. After the project, the Bosch Centre at IISc took over development by bringing together professors from other departments.
It also hired engineers and put together a development team. "We wanted to explore the field of data-driven robotics," said Bharadwaj Amrutur, professor of electrical engineering at IISc and chairman of the Robert Bosch Centre.
Data-driven robotics is a set of technologies that use data to get a robot to learn by itself. As the IISc project got off the ground, it was joined by Shishir Kolathaya from Georgia Tech University. Kolathaya, who has been working with walking robots since the undergraduate level, studied legged robots for his PhD.
When he joined, the Bosch Centre had a non-working prototype. The first real prototype - Stoch 1 - didn't carry batteries. Stoch 2 was twice as powerful, was designed to carry batteries and could walk for 15 minutes without being plugged into an electrical outlet. The third version will improve on looks and be even more powerful. The commercial prototype, when ready, will be bundled with an applica tion.
The Bosch team is mulling several applications - climbing coconut trees, doing surveys in difficult terrain, inspecting construction sites and so on. The project now has five engineers, apart from some faculty members. "We are planning to put a software development kit for people to programme," says Dhaivat Dholakiya, who is technical associate of the project.
Source: https://tech.economictimes.indiatimes.com/news/technology/iisc-engineers-take-fetch-to-a-whole-new-level/68418892 (Accessed on 16 March, 2019)
The robot is the size of a small dog with four legs and a thick but flexible spine. When connected to a battery, it starts walking on slender, articulated limbs, like a canine.
It even has a 'face' that looks vaguely like that of a pug. The developers at the Robert Bosch Centre for Cyber-physical Systems at the Indian Institute of Science (IISc) are calling it 'Stoch' and they say it's on target to become the country's first commercial "walking" robot.
Stoch has been under development since January last year. The first version was displayed a month ago at Aero India - it was heavy-footed and clumsy. A sleeker second version was developed just two weeks ago and a third will be ready in three months. A commercial variant is about a year away.
The robot uses machine learning to figure out how to walk by itself. Specifically, it uses reinforcement learning, where the machine learns over time to take the best possible action in return for the best reward.
After several million attempts - performed in computer simulation - ؙthe robot learns to walk. Some Indian Institutes of Technology (IITs) and IISc have been working on such devices in the past few years. There aren't any commercial versions in the market and no institution other than IISc has a prototype that uses reinforcement learning to teach the robot to walk.
The IISc project started when a student, Shounak Bhattacharya, did a master's project in the department of mechanical engineering. After the project, the Bosch Centre at IISc took over development by bringing together professors from other departments.
It also hired engineers and put together a development team. "We wanted to explore the field of data-driven robotics," said Bharadwaj Amrutur, professor of electrical engineering at IISc and chairman of the Robert Bosch Centre.
Data-driven robotics is a set of technologies that use data to get a robot to learn by itself. As the IISc project got off the ground, it was joined by Shishir Kolathaya from Georgia Tech University. Kolathaya, who has been working with walking robots since the undergraduate level, studied legged robots for his PhD.
When he joined, the Bosch Centre had a non-working prototype. The first real prototype - Stoch 1 - didn't carry batteries. Stoch 2 was twice as powerful, was designed to carry batteries and could walk for 15 minutes without being plugged into an electrical outlet. The third version will improve on looks and be even more powerful. The commercial prototype, when ready, will be bundled with an applica tion.
The Bosch team is mulling several applications - climbing coconut trees, doing surveys in difficult terrain, inspecting construction sites and so on. The project now has five engineers, apart from some faculty members. "We are planning to put a software development kit for people to programme," says Dhaivat Dholakiya, who is technical associate of the project.
Source: https://tech.economictimes.indiatimes.com/news/technology/iisc-engineers-take-fetch-to-a-whole-new-level/68418892 (Accessed on 16 March, 2019)
