Advances in artificial intelligence and situational awareness are set to underpin the next generation of collaborative robots, designed to safely work side by side with humans in industrial environments.
The industrial robotics market is expected to grow by 175% over the next decade, according to Loup Ventures research, and the primary focus of that growth will be on assistive platforms. A third of industrial robots sold by 2025 will be collaborative « cobots », designed to work safely alongside humans in factories and plants.
Cobots allow manufacturers to leverage the benefits of robots for performing tasks that are still too challenging for the robot to tackle alone, but there is still work to be done to strengthen the relationship between man and machine, says Dieter Fox, NVIDIA senior director of robotics research.
Cobots rely on cameras and other onboard sensors to monitor their environment, with advances in machine learning and machine vision set to improve their perception skills – not just when it comes to recognizing objects but also grasping concepts like context and intent.
By perception, I don’t just mean robots identifying things they can see through a camera, I also mean robots actually understanding their immediate environment. The ability for cobots to understand what is happening around them, including comprehending the actions of their human co-workers, will underpin their ability to predict what’s going to happen next. This is crucial for cobots in order to reach their full potential while keeping human safety a priority, something that can already be seen as one of the challenges that is holding back autonomous vehicles.
Machine learning will be a key tool in taking cobots to the next level and, as with autonomous vehicles, Fox believes cobots will benefit from training in high-fidelity virtual environments.
When it comes to real-world training, their improved perception skills will allow robots to learn from watching their human co-workers demonstrate complex tasks, perhaps whilst taking onboard additional spoken guidance. This kind of on-the-job training will be easier and more efficient than expecting the robot to simply memorize actions as a person physically moves them through the motions of a required task.
Once cobots are at work on the factory floor, a new generation of 3D sensors is required to calculate the distance between the robot and nearby humans in order to maintain a dynamic safety zone, says Dr Mohamad Bdiwi – head of the robotics department at Fraunhofer Institute for Machine Tools and Forming Technology (Fraunhofer IWU).
For cobots and humans to achieve optimal productivity, the robot’s perception will need to extend to facial detection. This will not just assist with identifying people and anticipating their actions, but will also allow cobots to better interpret important non-verbal communication such as facial expressions and gestures.
Verbal communication will be more important for domestic service robots, but it has limitations in a noisy industrial environment where important spoken commands could be lost or misinterpreted. Even the ability to comprehend simple and intuitive hand gestures, such as stop and go, is something we take for granted when communicating with other humans. Empowering cobots to read such non-verbal cues, when following instructions or assessing danger, will certainly enhance their collaborative abilities.
Interaction between humans and machines is a two-way street, with cobot designers also needing to allow for the emotional bond that people can form when they anthropomorphize robots, says Sari Nijssen – PhD candidate in human-robot interaction at Radboud University.
The more humanoid a robot appears, and the more it seems to display lifelike characteristics, the more humans care about its welfare, according to Nijssen’s research. She proposed a variation of the classic ‘trolley problem’ to participants: asking whether they would be prepared to endanger a robot in order to save several injured people. Participants’ answers varied according to how emotionally attached they were to the robot.
Evoking high levels of empathy from people might be useful when designing social robots for broader human interaction, but it could be counterproductive when it comes to designing industrial cobots. While a certain level of bonding is required for humans to accept their robotic co-workers, having too great a concern for the robot’s wellbeing might hamper overall productivity, Nijssen says.
Not much research has been done on long-term relations between people and robots, but it is an area that will require more attention as cobots become part of our lives. While cobots aren’t able to reciprocate empathy, they can perhaps learn to read interactions and simulate empathetic responses in order to better engage with humans – the key is finding the right balance to establish the best working relationship between humans and their cobot co-workers.