Resize Herbot post media to 70%

- Reduce video size from 100% to 70%
- Reduce web interface images from 100% to 70%
- Improve layout consistency with Agent Mcqueen post

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: leesg0107

_posts/project/2025-11-19-herbot.md

@@ -10,7 +10,7 @@ thumbnail-img: "/assets/img/Herbot/Herbot.jpeg"
 ---
 Herbot is an autonomous herb management robot I developed as part of the "AI Embedded Programming" course. The system autonomously handles herb cutting and removes diseased leaves.
 
-<video width="100%" controls>
+<video width="70%" controls>
   <source src="/assets/img/Herbot/Herbot-video.mp4" type="video/mp4">
   Your browser does not support the video tag.
 </video>
@@ -43,11 +43,11 @@ Hardware assembly was just the beginning. Each motor required careful calibratio
 
 ## Web Interface
 
-<img src="/assets/img/Herbot/Herbot-web.PNG" alt="Herbot Web Interface" style="width:100%;">
+<img src="/assets/img/Herbot/Herbot-web.PNG" alt="Herbot Web Interface" style="width:70%;">
 
 To make the system more accessible, I built a web interface for Herbot. Users can view all captured images, access system logs, and remotely control the robot's operations. The core functionality is the "scan" feature, which systematically photographs the plant from multiple angles. I trained two vision models—one for herb species identification and another for disease detection—on my local machine, then deployed them to run on the Raspberry Pi 4 accelerated by the Coral USB device.
 
-<img src="/assets/img/Herbot/Herbot-web-log.PNG" alt="Herbot Web Logs" style="width:100%;">
+<img src="/assets/img/Herbot/Herbot-web-log.PNG" alt="Herbot Web Logs" style="width:70%;">
 
 ## Vision Models
 
@@ -67,12 +67,10 @@ To achieve real-time performance on embedded hardware, I applied INT8 quantizati
 
 The optimization delivers substantial speedup—inference time drops from 80-150ms on CPU to just 10-15ms on the Edge TPU for Herbify, and from 70-120ms to 5-12ms for PlantDoc. This acceleration enables the robot to scan multiple plant angles efficiently, processing each captured image in under 30ms total for both species identification and disease detection.
 
-
-
 ## Conclusion
 
-This was an incredibly fun project, though not without its limitations. Budget constraints were significant—while support from my department helped immensely, I still lacked both sufficient funding and time. As a result, Herbot has no proper housing. I improvised the base and rotary stage using a car air freshener box, and attached the camera and gripper to the carriage with adhesive and electrical tape. It looks crude, but the cost-efficiency was unbeatable.
+This was an incredibly fun project, though not without its limitations. Budget constraints were significant while support from my department helped immensely, I still lacked both sufficient funding and time. As a result, Herbot has no proper housing. I improvised the base and rotary stage using a car air freshener box, and attached the camera and gripper to the carriage with adhesive and electrical tape. It looks crude, but the cost-efficiency was unbeatable.
 
-This project taught me a critical lesson about cost-effectiveness in engineering. Most university coursework exists in an unrealistic bubble where practical constraints like budget and robustness simply don't matter. You're not building products for real users or dealing with real-world tradeoffs. But Herbot forced me to work within severe limitations—both financial and computational—and that made me think deeply about engineering compromises.
+This project taught me a critical lesson about cost-effectiveness in engineering. Most university coursework exists in an unrealistic bubble where practical constraints like budget and robustness simply don't matter. You're not building products for real users or dealing with real-world tradeoffs. But Herbot forced me to work within severe limitations(both financial and computational) and that made me think deeply about engineering compromises.
 
-I realize now that this inexperience with real-world constraints was why my first drone project, Soltrone v1, failed. Back then, I naively thought I could just mount a flight controller, four propellers, and a Raspberry Pi together to create an autonomous drone. Reality hit hard with countless practical problems I hadn't anticipated. But the lessons from Herbot—understanding budget constraints, optimizing for limited resources, and making smart tradeoffs—will be invaluable for Soltrone v2.
+I realize now that this inexperience with real-world constraints was why my first drone project, Soltrone v1, failed. Back then, I naively thought I could just mount a flight controller, four propellers, and a Raspberry Pi together to create an autonomous drone. Reality hit hard with countless practical problems I hadn't anticipated. But the lessons from Herbot(understanding budget constraints, optimizing for limited resources, and making smart tradeoffs)will be invaluable for Soltrone v2.