The project is aimed towards facilitating autonomy to an existing warehouse electric Pallet Jack. The vehicle used is meant to be manually operated. Thus, as a first step the necessary hardware and software changes for the drive-by-wire system were incorporated. To control the pallet jack a kinematic model was used which was also validated using a motion capture system (VICON). Three trajectory controllers namely Model Predictive Controller (MPC), Trajectory Optimised Linear Quadratic Regulator (T-LQR) and PID were implemented and compared under full state feedback from VICON. Finally a complete pallet DOCKING-PICKUP-DROPOFF (DPD) was demonstrated using a pure pursuit based docking behaviour. The future works include achieving the same functionality using just the onboard sensors like LIDAR, stereo camera and upright camera, for robust low cost indoor autonomy. Details of the implementation are of proprietary nature and hence the code can't be shared.

As a next step towards platform independent indoor autonomy , we demonstrate Docking with a pallet using just the state data acquired from Intel RealSense Tracking Camera. The camera tracks features in the environment and use that to ascertain its pose information. Its robust to fast motions and provides a good state estimation. The video shows successfull docking with the pallet with a fair degree of accuracy.

The above demonstration showcases the ability of our pallet jack to effectively navigate between waypoints. In this case there is sufficient buffer given to the system to determine the position of the marker which decides the next waypoint. The Pallet Jack uses MPC to effectively plan a trajectory between waypoints. Only when the previous waypoint is reached , the next waypoint is followed. The state feedback here is obtained using a Motion Capture System.

Inspired by the Motion capture system at the lab I am working, I very much wanted to have a similar system at my home for easy experimentation. Because the VICON motion capture system is pretty expensive I utilised what I already had i.e. an android phone. Using simple computer vision techniques and making use of Open-cv I ended up making an indoor perception system called VISPER. I utilised it to track my differential drive robot shiro in real-time.