Mobile Manipulation Hackathon (IROS 2018) - TAMS Team
The TAMS Team presents PALRobotics’ mobile manipulation hackathon demo at IROS 2018, showing a robot serving a customer at a bar.
In this talk
- 0:09
Detecting a customer and taking an order
The robot is meant to notice a person behind the bar, approach, and find a menu. It combines face recognition with depth-image clustering to detect the person, then uses camera images and finger-tip detection to identify a selected menu item. When it cannot find the menu, it prompts the user to try again.
- 2:11
Planning motions and identifying bottles
After recognizing an order, the robot looks up a recipe and pours its ingredients in sequence; the team demonstrates with water. Its MoveIt-based manipulation uses motion primitives for approaching, grasping, lifting, and moving an object. Bottle identification combines tabletop cluster segmentation, a convolutional neural network trained on about 1,000 images, and cylinder fitting with RANSAC.
- 3:23
Pouring while keeping the bottle upright
The robot picks up a bottle and moves it into a previously defined transport pose while maintaining an upright constraint. For pouring, the team uses a hard-coded Cartesian motion relative to the glass center and executes it only when the planned motion is expected to preserve that constraint.
- 5:46
Finding the glass and placing the bottle
The team detects the glass using the camera’s infrared image, applying high-pass and median filtering to estimate its bottom and overall shape. The method has limited resolution and may not be fully accurate. The demo also exposes navigation and placement limitations: the robot reports reaching the table while still about 50 centimeters away, and the team says it did not implement precise object placement.
- 7:11
Lessons from a one-week build
The team had one week with a robot they had not seen before. It relies on an approximate local table position and default segmentation for placement rather than explicitly detecting the tabletop surface. A last-minute change failed, and localization problems remained as the robot returned to idle.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.