Flipside VLA-AV Project VLA-AV Sidecar Tool (POC Version)

---

This project organizes Flipside’s work on Vision-Language-Action (VLA) annotation for Autonomous Vehicles (AV).

The project supports workflows similar to Kognic’s reasoning-based annotation format, where annotators:

•  Review model-generated scene facts
•  Review model reasoning about the ego vehicle
• Correct errors or omissions
• Justify the ego vehicle’s action 

 A key supporting component is the Flipside Language Sidecar, a lightweight assistant tool that helps annotators produce clear and consistent English reasoning.

The Flipside Sidecar assists annotators who think in:  

• Tagalog 
• Bisaya 
• Taglish
• English  

to produce clear English annotations that follow the VLA workflow.  

The tool performs language normalization and formatting only.  

It does not create annotations and does not interpret the scene. 

Core Responsibilities 

The Sidecar: 

•  Normalizes mixed language into clear English
• Compresses reasoning into concise statements 
• Applies standardized driving terminology
• Formats reasoning according to the VLA structure  

The Sidecar must never introduce new scene facts.

Initial deployment is limited to the current VLA-AV Proof-of-Concept (POC) with Kognic.

Objectives of the POC phase:

• Improve annotator language consistency
• Reduce reasoning writing time
• Evaluate annotation productivity impact
• Assess whether the tool should become a permanent internal platform