I competed in a design competition with my friends Chris Lansdale, Riddhiman Roy, and Rana Abdelrahman where we had to automate the warehouse operations of a store by creating a storage and retrieval robot.
The first thing we did was ideate methods to solve the problem to see if we could come up with new and unconventional approaches.
The cost-benefit analysis of different designs.
Ultimately, we decided upon an automated robot that navigates through the warehouse using Cameras and LiDAR. Every item in the warehouse would be put in trays and have QR codes associated with the front of each tray. The robot would be connected to a shared database containing the inventory and location of various items around the store and would use SLAM to localize itself and the items of interest. Ultimately, the trays would be lifted using a forklift-like mechanism attached to the robot. Since the emphasis of this designathon was on mechanical and electrical design as opposed to software, the following shows the process we followed to CAD and create a downsized prototype of the concept.
The CAD of the robot we came up with. CAD models were made by Chris Lansdale.
Below is a schematic of the design highlighting a few key design choices. First, the reason a forklift design to pick up trays was opted for was to simplify the problem of how to grasp and carry items of various shapes and sizes. Since the warehouse is a controlled environment, enforcing a policy where stored items are kept in trays and trays instead of objects are retrieved can work better. In the design space we were allowed to assume we have one employee in this warehouse. We decided to use this employee for retrieving the specific item that needs to be sent out from the tray brought by the robot. Next, the use of SLAM and object detection using AI is a logical choice since the warehouse space is contained and can easily be mapped and updated with every aisle the robot traverses. This SLAM would only be used for global navigation within the warehouse while AI based object detection identifies obstacles and would be used for local trajectory planning. As for path and task planning, that would be done by solving an optimization problem to minimize distance travelled based on queried items from a database. Finally, the use of omnidirectional wheels was provided since the space in a warehouse can be constrained and this enables the robot to have all-direction control. These features are visualized below.
Schematic of storage and retrieval robot.
Riddhiman and I prototyped various aspects of the robot by building a downsized model of the robot. First we began by building the circuit and testing the wheels.
Circuit to power the motors.
Next, we built the motorized pulley mechanism which would control the forklift. We tested the forklift's capacity to bear loads and realized that a cross hatch patterned support below the forklift bars would help improve the load bearing capacity of the forklift arms.
Building the motorized pulley ato control the forklift arms nd testing it's ability to lift objects
In the end, we had a prototype of a robot that could lift objects up and place them down. There wasn't enough time in the designathon to also prototype a perception and planning system of this robot. It would be interesting to see whether the forklift arms would easily find their way under the trays since the initial assumption is that would be an easier problem to solve than to build an intricate mechanism to grab and lift objects or boxes.
If you are interested in learning a bit more about the design process we followed and some other aspects of the project, look at this presentation that Rana helped us make which we presented.
Demonstration of robotic movement