Armtender is a bartending project built using a LeRobot arm to deliver drinks in a fun way. It spins a fortune wheel with a button. Once the button is pressed, a green light lits up and it looks for the drink the fortune wheel landed on. Based on the arrow location, it decides for what drink to pick up. Then it goes to the drink, picks it up and delivers the drink can through the pick-up window.

Team number: 16
Team name: Remphia
Team members:

• Emme McDonald (@giragon6)
  
• Ruzanna Gaboyan (@xsollwa)
  
• Afia Ava (@afia-ava)

Armtender demonstrates how, in the future, both luck-based games and food and beverage service can be automated. In the case of the former, dealer bias can be reduced to increase fairness of played games. For the latter, more precision and reduced contamination risk can be achieved.

Our approach is unique in that it is end-to-end automated: it does not use sensors to tell the robot which drink to grab. The robot instead learns that it should grab a particular item when the wheel's arrow is over that item's color. It also incorporates additional hardware in the wheel of fortune. which allows for a degree of randomness in how long the wheel spins. This increases the complexity of the robot's task but more accurately reflects real-life use cases.

Training
Number of Episodes: 1000
Batch size: 1288
Number of Steps: 10kk

Our implementation is specific to our game and environment (i.e., a wheel with color-coordinated sections and four items positioned consistently). However, our multi-stage methodology with increased emphasis on robot inference over sensor feedback is adaptable to numerous applications. Our robot is activated by using the lerobot-record command, which pulls from our trained model to begin an inference episode.

• Watch the teleoperation demo