Project Management Review

Team G

Contents

• Project Description
• Subsystem Status
• Key Challenges and Associated Plans
• Project Management method
• Project management tools and Techniques
• Lessons and Improvements
• Schedule and milestone
• Risk Management
• Budget

​

Project Description

Our goal is to develop a robot based on the Hello Stretch RE1 platform that can harvest tomatoes.

Our robotic system will:

1. Make use of custom-designed tool.
2. Autonomously navigate around the controlled setup environment that mimics a vertical farm
3. Identify the interaction sites where the task must be performed
4. Perform the desired harvesting action.

Subsystems status (Test Environment)

• New plant stems made from floral wire
• Foliage created using fake vine leaves to recreate same noisy backdrop
• Reduces need for upkeep

​

Subsystems status (Perception)

Using Faster RCNN for tomato detection - Performance:

• Detection mAP=0.53
• Detection rate of 1~2 FPS (very slow)

​

Future plans:

• Transferring from Faster RCNN to YOLO V5
• Detection mAP=0.59
• Expecting improved inference time
• Planning to optimize the model further using TensorRT

​

Subsystems status (Manipulation)

• End Effector reaches within +/- 10 cm of the required goal.
• Only arm translation and lift are engaged

Future Plans:

• Include base movement to reach a 3D point.

​

​

Subsystems status (Navigation)

• Currently our stack uses a high-level planner that works in tandem with a pot-detection algorithm.
• The high-level planner uses these guides to plan a route through the farm. Intermediate goal way-points are passed to the movebase package in ROS for tracking.

Future plans:

• Implement our own custom planner which makes use of the pot-detection algorithm directly.
• Develop our own pure-pursuit controller to eliminate reliance on movebase package.

Subsystems status (Tool Design)

• Current tool used is harvester - consists of cutter and gripper.

​

• Two alternatives being evaluated - To continue with both cutter and gripper, or to simply stick with gripper alone.

​

Key challenges & associated plans

• The platform is proving to be more problematic than anticipated. We hope to work in tandem with the Hello Robot team to address these issues.

​

• We are trying to devise approaches that are independent of position control of the arm and rely instead on visual feedback since the arm itself is unreliable.

​

• Instead of relying on the low-level navigation controls from MoveBase we hope to use our own pure-pursuit controller to navigate across the farm allowing better control and integration with the rest of the sub-systems.

​

Project Management

Shift from Traditional to Agile

DISCOVER

DESIGN

DEVELOP

LAUNCH

Project Management tools and techniques

Weekly Monday standups

Bi weekly sponsor meetings

All hands Friday sessions

Slack to maintain updates

Jira to keep track of sprints

Lessons & improvements

Lessons:

• Technical : Did not use PM tools efficiently
• Personnel: Lack of Communication
• Scheduling : Integrated testing for harvesting and pollination started late
• Materials: Last minute problems with the robot and its accessories
• Testing: interface and communication between the subsystems

​

Improvements :

• Regular meetings with stretch support
• Quicker iterations
• More teamwork, more collaborations and standups
• Vetting personal/sub-system level goals before the whole team
• Test runs every Friday
• Using Jira and slack more efficiently

Schedule/milestones

High level goals for Fall semester

1. Base orientation to reach a 3D point
2. No reliance on movebase for navigation
3. Improve detection model’s inference time on Jetson
4. Tool performs needed action 100% of the time

Risk management

Before Mitigation Plan

After Mitigation Plan

Budget

Questions