Schedules:

• Progress Review 1

Test Plans:

• Spring Test Plan
• A general spring schedule is shown below:

Spring Performance Results:

• Button Detection:
  • Requirement: SR > 33%.
  • Achieved: 90% (18/20) for both in and out the elevator.
• Press Elevator Button:
  • Requirement: SR > 33%.
  • Achieved: 85% (17/20) for raised button. 50% (5/10) for inset button.
• Visual Servo Base:
  • Requirement: SR > 33%.
  • Achieved: Detection: 82% (36/44), Servoing: >95% (19/20), Combined 78%
• Door State Detection:
  • Requirement: SR > 33%.
  • Achieved: ~100% (3/3), generalized to different elevators
• Enter Elevator:
  • Requirement: SR > 33%.
  • Achieved: ~100% (20/20)
• Elevator Floor Detection:
  • Requirement: SR > 33%.
  • Achieved: ~80% (4 out of 5 elevators, 100% on the 4 elevators)
• Exit Elevator:
  • Requirement: SR > 33%.
  • Achieved: ~100% (6/6)

Part List:

• Track of Hardware

Risk Management: Identification of major risks

we have identified and continuously monitored key risks that could affect the progress of our project. These risks span both technical and logistics domains. The risk matrices outline each identified risk along with its likelihood, severity, and planned mitigation approach. In general, our mitigation are mainly designed to reduce the likelihood of all the risks identified. The majority of the risk will be under control as the project proceeds as planned. The risks are listed below.

Design Brain-Storming:

• Each module outputs a confidence score, and the behavior tree decides whether to execute, retry, change viewpoint, or switch to a fallback route.
• Store doors, elevators, stairs, and buttons as semantic landmarks with preferred approach poses, viewpoints, and interaction poses.
• Use high-level symbolic planning for route/task sequencing, but use local reactive controllers for fragile interactions like door opening and button pressing.
• Start with VLM-based object localization, then switch to lightweight tracking or geometric priors during execution to reduce latency and improve robustness.
• Design every major skill with explicit failure modes, timeout logic, retry limits, and user-facing explanations.
• First validate isolated artifact skills, then short skill chains, and only finally the full multi-floor mission.