Summary of Key Points
DQ and the AI robotics company Sharpa have opened a 24-hour robotic ice cream shop in Shanghai, marking a significant step from robotic demonstrations to real-world commercial applications. The robots use human-made equipment to produce ice cream, following a 55-step process that includes the iconic DQ skill of pouring ice cream without spilling. However, the robots are slower than humans (6 minutes per cup compared to 2-3 minutes). For now, the shop offers only Oreo-flavored ice cream and operates on a dual-system: robots handle scheduled orders, while humans handle walk-in requests. The short-term goal for both parties is not to make a profit but to thoroughly understand the business dynamics—accumulating real data and improving the robots’ adaptability. This will lay the foundation for future scalability in other fast-food categories such as hamburgers and pizzas. The long-term aim is to make the robots as efficient as humans and reduce costs, thereby achieving a positive return on investment.
I. Practical Performance of the Robots: Slow but Stable, Still in the “Learning Phase”
The robots’ operational process is comprehensive: picking up the cup, making the ice cream, adding toppings, mixing, and pouring it without spilling. Although they are slower (6 minutes compared to 2-3 minutes for skilled humans), the quality of the ice cream is consistent, with only minor issues like a few crumbs falling off when pouring. Currently, the robots handle only scheduled Oreo-flavored orders, with walk-in customers being served by humans at the same price.
Why limit the orders to scheduled ones? The robots are not yet fully proficient, and scheduling ensures they work at a steady pace, avoiding the chaos of handling many walk-in requests simultaneously. Observations for an hour showed that the robots handled the scheduled orders with ease, but they might struggle with a higher volume of customers.
II. Technical Challenges: Beyond Pre-set Actions, Handling Real-World “Accidents”
The robots do not use specialized equipment; instead, they use human-made ice cream machines, cups, and spoons. This presents difficulties in terms of “perception” and “adjustment”:
- Force and Touch Control: The robots need to grasp the cups with the right amount of force (too much force may result in excess ice cream, too little force may cause slipping). They adjust their grip based on tactile sensors and AI models.
- Real-World Issues: For example, the temperature of the ice cream may rise after the machine has been running for a while, affecting its consistency and mixing. These issues cannot be simulated in the lab and can only be addressed through actual operations and data collection to update the models.
It took several months of training for the teams to train the robots to complete the 55-step process, using real machines for remote operations, simulation, and human operation videos.
III. The Short-term Sacrifice of Profit: Focusing on Long-term Value
The initial investment in opening this shop was substantial, including converting an existing hamburger restaurant into a robotic operation area and covering the costs of the robots and training. With the current order volume and efficiency, a profit is not likely in the short term. However, the motivations for both parties are different:
- DQ’s Approach: To embrace innovative technology and see if robots can be integrated into existing store processes without the need for equipment changes or standard adjustments.
- Sharpa’s Vision: To fully adapt the robots to real commercial environments, such as equipment wear and temperature variations, and accumulate sufficient real data. This will enhance the robots’ generalization ability, allowing them to be easily applied to other food categories.
Li Yifan, co-founder of Sharpa, emphasizes that the potential value of fully integrating a robot into a business process can be substantial—for instance, if the success with ice cream leads to the expansion into other fast-food items, the costs can be reduced, making it profitable.
IV. Future Opportunities and Challenges: Scaling Depends on Efficiency and Cost Reduction
For robotic restaurants to be profitable in the long term, three issues need to be addressed:
1. Speed: The current 6-minute speed is too slow; only when the robots match human speed will they create comparable value.
2. Equipment Durability: The robots need to be durable, similar to cars, to reduce the cost per use.
3. Cost Reduction: The current high cost of robots (in the hundreds of thousands of yuan) needs to be lowered to make them more cost-effective.
The companies plan to have the robots operate during off-peak hours and gradually increase their usage in the store. However, with the onset of autumn in Shanghai, ice cream sales will decrease, posing new challenges to the robots’ efficiency and the store’s profitability.
Li Yifan also warns against creating artificial roles just for the sake of using robots, such as dedicated personnel to fix malfunctions. The real goal is for robots to solve practical problems, such as addressing staffing shortages at night or high labor costs.
Conclusion
Robots making ice cream are not just for show; they are being trained for real-world applications. Although they are slow and expensive at present, they could become a valuable tool for the fast-food industry to reduce costs and increase efficiency. Whether they can become a necessity depends on the balance of efficiency, cost, and market demand.