This page gives a brief video story of the BluPe’s progress.

📹 Progress Videos

🗓 Jul 8, 2025

Blackberries !! Blackberries !! Blackberries!! Bigger & Softer. Probably the biggest reason for shifting to blackberries is that there is a significantly large portion in the US of blueberries that are picked by a harvester, while for blackberries basically all fresh market berries are picked by hand as they are more delicate. Since softness was not an issue for blueberries, and BluPe was able to adjust to different sizes of berries Blackberries seemed as a better starting fruit. Mechanically they also grow individually rather than in clusters, and are larger - easier to clearly spot in an image. For the demo 900 picks were done on a white background and 150 with leaves. Still there is room for improvement with initial motions.

🗓 Apr 30, 2025

Given the previous datasets collected, and roughly 100 more outdoor demonstrations, BluPe was able to pick 3 berries, this being the first one. Special thanks to Clay Ranch Berry Farm. While most berries grow in clusters, we can start by picking those that are growing individaully, and once that is refined, the arm will learn to pick from clusters more easily.

🗓 Apr 14, 2025

Arm is bigger, stronger, and we see that we are able to pick berries from 3D space in a noisy environment. Using larger arm is harder as arm needs to be especially persise with final movementes, where minimal movements in the base motors have a large effect on end gripper position as the arm is larger. While this video still took many attempts, we can see that it successfully picks, even with a bunch of leaves.

🗓 Jan 7, 2025

This video was unrealistic, yet mannaged to paint a vision. While a policy was trained, the berry was placed basically in the only spot where the arm could collect. Later we had to figure out how to collect data so that the arm could generalize and pick berries more robustly.