- Home » News » Technology News
Voice-controlled robot programming is ‘90% faster’
Mitsubishi Electric has developed a technology that, it claims, will allow industrial robots to perform tasks such as sorting and arranging, as quickly as humans, without needing to be taught by specialists. It says that the technology – which uses speech recognition to issue voice instructions to robots – will allow them to be programmed and fine-tuned in 10% or less of the time needed for conventional approaches.
Mitsubishi expects the technology to be particularly useful in facilities such as food-processing plants where products change frequently, making it difficult to use robots.
The technology allows robots to program and adjust their movements in response to simple spoken or menu commands given by non-specialist operators. Mitsubishi claims to have developed the first voice-instruction AI for industrial robots that recognises spoken commands accurately, even in noisy factory environments. In a food-processing plant, for example, an operator could tell a robot to “pack three pieces of chicken in the first section of the lunch box.”
After receiving the spoken or menu commands, the system scans the working area using a 3D sensor and programs the robot’s movements automatically. The operator can fine-tune the movements using further commands.
If a voice instruction is ambiguous – such as “a little more to the right" – the AI can infer implied meanings. Alternatively, menus on a tablet can be used to issue instructions or to select categories such as “where,” “what” and “how many” to generate simple programs.
3D sensors detect information about the work area – including images and distances – which is processed using augmented reality (AR) to create 3D simulations that operators tcan view on tablets to confirm that their instructions will have the intended results. The system can also can recommend the ideal positioning of a robot in a virtual AR space without needing a dedicated marker – claimed to be another industry first.
By allowing robots to program their own movements, including avoiding obstacles, the system reduces the work traditionally associated with gathering information about an operating environment, inputting data and confirming operations using simulators and/or the actual equipment.
The system can complete these processes in 10% or less of the time needed using conventional methods. In a typical application, says Mitsubishi, inputting information about the operating environment can be cut from 16 hours to 30 minutes, programming can be reduced from four hours to 30 minutes, and subsequent adjustments and modifications can be slashed from 40 hours to four. The means that time taken to get the application up and running could plummet from 60 hours to just five.
Mitsubishi expects its system to be particularly useful for automating processes that are not suited to robots, such as food processing, where products change frequently, requiring robot programs to be updated for each change.
Conventionally, increasing a robot’s operating speed requires time to achieve the desired trajectories because operators must use simulators and/or the actual robot to determine the best conditions. For its new system, Mitsubishi has developed a trajectory-generation technology that optimises robot movements using information about surrounding obstacles and the performance of the robot.
The company has also developed a technology that optimises acceleration and deceleration rates for robot arms to generate velocity patterns that achieve the shortest arm-travel time based on the range of forces that the robot can exert. This, combined with the trajectory-generation technology, helps to cut the time needed to adjust the robot’s movements.
In another time-saving measure, the system adjusts and optimises the opening and closing of robot grippers. Conventionally, such adjustments are performed manually using simulations. The new technology adjusts and optimises the timing automatically depending on the characteristics of the gripper and the object being grasped, thus eliminating time-consuming manual adjustments, while improving efficiency.
In addition, the grasping-recognition AI, combined with a 3D sensor located above the application, rapidly determines the best position for grasping, thus saving more time. There is also a camera attached to the gripper that allows the robot to self-correct its movements if its position, or that of its target, changes.
By optimising robot arm and gripper movements, the new technology is said to allow robots to work as fast as humans, taking a minimum of 2s to grasp an item and place it in a designated location.
To simplify the process of adding customised functions, the system supports the Robot Operating System (ROS) platform. Mitsubishi has developed a function that links ROS to Edegecross – a Japanese open software platform for edge computing that combines factory automation with IT. The function will allow users to manage information on Edgecross, ROS and entire systems. The Edgecross connections and multi-vendor capabilities will simplify the monitoring and analysis of entire production lines to enhance productivity and quality.
Mitsubishi is aiming to commercialise the new technology in or after 2023, following further performance enhancements and testing.
Mitsubishi Electric: LinkedIn Facebook
