After researching, our engineers combined Dobot’s cobots with the most popular technology this year, ChatGPT, and applied it in the real world.
It has more powerful language understanding and text generation ability which can ‘understand’ human words and communicate just like mankind, with the ability of even writing papers, scripts, and code.
Robot Bar Tender
A good bartender should be good at communication by providing the guests a certain amount of comfort at the right time. In this project, the engineers incorporated ChatGPT into the project and transformed cold robot into an excellent bartender with a warm heart.
We trained ChatGPT in advance, telling it in detail the role it should play (bartender), the recipes of various cocktails, the taste and so on.
After training, the robot bartender becomes reliable and attentive with more autonomous judgment. For example, if a customer makes an unreasonable request for tea in a Long Island Iced Tea cocktail, ChatGPT will explain that there is no tea in the Long Island Iced Tea, but a mix of spirits and coca cola instead.
ChatGPT will analyze and judge by itself according to the customer’s current emotional condition and recommend different drinks to cheer them up. It will then presume customers’ needs to adjust based on the context which means it is better than usual AI where usually respond with ‘I don’t know what you mean’.
More importantly, after the cobot grants control interface information, trajectory tracking and codes will be produced for drink making based on the characteristics of different cocktails.
The bar tender can daintily make all kinds of cocktails with DOBOT CR3 and CR5 cobots.
Smart Object Sorting
In the past, engineers needed to enter a large amount of code for operation, and different robots might involve different programming languages. With the help of ChatGPT and AI technology, engineers can directly describe what they want to do verbally, then it automatically compiles into machine language which are perceptive to fully engage the robot's hands, eyes, and brain for various actions.
For a challenge, we asked the robot to grab a pound of oranges from a variety of items, automation allows ChatGPT to finish a series of tasks like fruit identification, fruit grabbing, and weight calculation simultaneously to complete the task.
When asked to classify items on the table, ChatGPT identifies the items by itself with automation carried out using a camera which automatically complete the classification task. This increases productivity as it saves time on that relied on manual labeling and classification.
From the two practical cases, ChatGPT can indeed interact by understanding and learning human language and combining it with reality. It can also quickly generate robot codes that match different scenarios and user’s needs, reducing development time and cost significantly.
From the technical level, the ChatGPT model learns human a priori knowledge and incorporates all kinds of physical perception. By using collaborative robots as the carrier, enhancing the experience of human-machine interaction and collaboration, robots become smarter and humanized. With this foundation, Dobot has formed its own technical framework for large-scale robot modeling.
In the future, Dobot will continue to innovate and breakthrough in the field of ‘AI + robot’, by colliding with more cutting-edge technologies for collaborative robots to have autonomous perception, smart decision-making and fine operation. This can help promote the application of large-scale models in advanced robot manufacturing and commercial services.
