ROS2 Real-Time Boundary Detection | Waveshare UGV Beast Bot
Budget / Salary₹12,500–37,500
TypeFreelance project
LocationRemote
Posted1 hour ago
I need help bringing my UGV Beast back to full sensing capability and then pushing it further: I want every object the robot sees to be ring-fenced by a crisp, real-time outline that I can open from any laptop or phone browser on the same network.
Location: Bangalore, India (preferred to work from office where the Hardware is located)
Current hardware already on board
• Raspberry Pi running ROS 2 Humble (we have purchased a new module, as existing RPi seems to have issues, troubleshooting would be required)
• Luxonis OAK-D Lite on a small pan-tilt head
• LD19 360° LiDAR
No other sensors, lights or emitters may be added; the solution must rely entirely on those three devices.(we already have lights in the bot) I have some ROS 2 nodes running, but depth frames from the OAK-D Lite no longer publish correctly and the LiDAR scan topic is intermittent. Step one is to restore clean, time-synced data streams.
Real-time boundary detection requirements
• Fuse stereo depth with LiDAR: in normal lighting the outline should be driven by depth; in total darkness fall back to LiDAR where depth confidence drops.
• Draw visible contours around any object class—people, furniture, random clutter—no semantic labelling needed, just the edges.
• Stream the annotated video (≈10–15 fps is fine) to a lightweight web page served directly from the Pi;
• Maintain low-latency control of the pan-tilt so the camera view remains stable while detection overlays update.
Acceptance criteria
1. `oakd_rgb_depth`, `oakd_depth_aligned`, `ld19_scan` topics publish at stable, matching timestamps.
2. Web page automatically loads the annotated stream and shows outlines at ≤150 ms end-to-end latency on a local Wi-Fi network.
3. In a fully dark room, the display falls back to LiDAR-driven contours wherever stereo depth confidence < threshold, with a brief on-screen indicator that fallback is active.
4. Provide all ROS 2 packages, launch files and a short README so I can reproduce the setup on a fresh SD card.
If this matches your ROS 2 and computer-vision toolbox (OpenCV, DepthAI, PCL or similar), let’s talk timelines and milestones.
Location: Bangalore, India (preferred to work from office where the Hardware is located)
Robot spec:
https://robocraze.com/products/waveshare-ugv-beast-open-source-off-road-ai-robot-chassis-kit-with-raspberry-pi-4b-and-pan-tilt-pt-pi4b-ros2?variant=47125905146080&country=IN¤cy=INR&utm_medium=product_sync&utm_source=google&utm_content=sag_organic&utm_campaign=sag_organic&srsltid=AYCX9vKnW_aV1HceOH5eEslQJjRo4Bxden-ZGz8d-3grURHSm-GzNoZp-qE&gclid=CjwKCAiAhqHXBhAXEiwATbksU2Cnzg2o-mUpx2qCLwqAYt4-367Iv96vGq4DGS66qMD6lvMP3zI_8BoCm4gQAvD_BwE
Location: Bangalore, India (preferred to work from office where the Hardware is located)
Current hardware already on board
• Raspberry Pi running ROS 2 Humble (we have purchased a new module, as existing RPi seems to have issues, troubleshooting would be required)
• Luxonis OAK-D Lite on a small pan-tilt head
• LD19 360° LiDAR
No other sensors, lights or emitters may be added; the solution must rely entirely on those three devices.(we already have lights in the bot) I have some ROS 2 nodes running, but depth frames from the OAK-D Lite no longer publish correctly and the LiDAR scan topic is intermittent. Step one is to restore clean, time-synced data streams.
Real-time boundary detection requirements
• Fuse stereo depth with LiDAR: in normal lighting the outline should be driven by depth; in total darkness fall back to LiDAR where depth confidence drops.
• Draw visible contours around any object class—people, furniture, random clutter—no semantic labelling needed, just the edges.
• Stream the annotated video (≈10–15 fps is fine) to a lightweight web page served directly from the Pi;
• Maintain low-latency control of the pan-tilt so the camera view remains stable while detection overlays update.
Acceptance criteria
1. `oakd_rgb_depth`, `oakd_depth_aligned`, `ld19_scan` topics publish at stable, matching timestamps.
2. Web page automatically loads the annotated stream and shows outlines at ≤150 ms end-to-end latency on a local Wi-Fi network.
3. In a fully dark room, the display falls back to LiDAR-driven contours wherever stereo depth confidence < threshold, with a brief on-screen indicator that fallback is active.
4. Provide all ROS 2 packages, launch files and a short README so I can reproduce the setup on a fresh SD card.
If this matches your ROS 2 and computer-vision toolbox (OpenCV, DepthAI, PCL or similar), let’s talk timelines and milestones.
Location: Bangalore, India (preferred to work from office where the Hardware is located)
Robot spec:
https://robocraze.com/products/waveshare-ugv-beast-open-source-off-road-ai-robot-chassis-kit-with-raspberry-pi-4b-and-pan-tilt-pt-pi4b-ros2?variant=47125905146080&country=IN¤cy=INR&utm_medium=product_sync&utm_source=google&utm_content=sag_organic&utm_campaign=sag_organic&srsltid=AYCX9vKnW_aV1HceOH5eEslQJjRo4Bxden-ZGz8d-3grURHSm-GzNoZp-qE&gclid=CjwKCAiAhqHXBhAXEiwATbksU2Cnzg2o-mUpx2qCLwqAYt4-367Iv96vGq4DGS66qMD6lvMP3zI_8BoCm4gQAvD_BwE
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.