Uploaded March 2026 | Updated November 2026, 14 hours ago
In this video, we chat with Vishrut Kaushik, Senior Robotics Engineer at Peer Robotics, to discuss how they are using computer vision to build collaborative autonomous mobile robots (AMRs) that move heavy pallets and trolleys in industrial environments. Vishrut explores how computer vision unlocks new possibilities for giving robots the context needed to seamlessly work alongside humans.
We also dive into the technical hurdles of training custom vision models to handle extreme edge cases, such as detecting worn-out wooden pallets in pitch-black trailers or identifying assets completely covered in reflective stretch wrap. Vishrut explains how his team built a continuous active learning loop using Roboflow APIs to capture failed detections from the robots, re-train their datasets, and push fine-tuned model updates back to the factory floor in just a day.
= Additional Resources =
Roboflow: roboflow.com/industries/robotics
Peer Robotics: peerrobotics.ai
= Chapters =
00:00 - Introduction to Peer Robotics & Collaborative AMRs
03:36 - Demo: Autonomous Pallet Unloading & Outbound Operations
09:01 - The Role of LiDAR vs. Computer Vision in Robotics
14:09 - Managing Unpredictable and Chaotic Warehouse Environments
17:57 - Edge Cases: Dark Environments and Stretch-Wrapped Pallets
23:28 - Vision Models: Oriented Bounding Boxes and Instance Segmentation
29:10 - The Future: Semantic Scene Understanding and Safety Monitoring
31:49 - Connecting the Robot to Roboflow APIs for Continuous Learning
33:53 - Career Advice: Obsess Over the Problem Statement
39:33 - Onboard Hardware: Local CPU and GPU Processing
42:38 - Deployment Timelines: From Hours to Weeks
44:33 - Accelerating Model Training with Roboflow
In this video, we chat with Vishrut Kaushik, Senior Robotics Engineer at Peer Robotics, to discuss how they are using computer vision to build collaborative autonomous mobile robots (AMRs) that move heavy pallets and trolleys in industrial environments. Vishrut explores how computer vision unlocks new possibilities for giving robots the context needed to seamlessly work alongside humans.
We also dive into the technical hurdles of training custom vision models to handle extreme edge cases, such as detecting worn-out wooden pallets in pitch-black trailers or identifying assets completely covered in reflective stretch wrap. Vishrut explains how his team built a continuous active learning loop using Roboflow APIs to capture failed detections from the robots, re-train their datasets, and push fine-tuned model updates back to the factory floor in just a day.
= Additional Resources =
Roboflow: roboflow.com/industries/robotics
Peer Robotics: peerrobotics.ai
= Chapters =
00:00 - Introduction to Peer Robotics & Collaborative AMRs
03:36 - Demo: Autonomous Pallet Unloading & Outbound Operations
09:01 - The Role of LiDAR vs. Computer Vision in Robotics
14:09 - Managing Unpredictable and Chaotic Warehouse Environments
17:57 - Edge Cases: Dark Environments and Stretch-Wrapped Pallets
23:28 - Vision Models: Oriented Bounding Boxes and Instance Segmentation
29:10 - The Future: Semantic Scene Understanding and Safety Monitoring
31:49 - Connecting the Robot to Roboflow APIs for Continuous Learning
33:53 - Career Advice: Obsess Over the Problem Statement
39:33 - Onboard Hardware: Local CPU and GPU Processing
42:38 - Deployment Timelines: From Hours to Weeks
44:33 - Accelerating Model Training with Roboflow










