IdeaWell

Project ideas from Hacker News discussions.

The Waymo World Model

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

1. Waymo’s “world‑model” and the LIDAR‑vs‑camera debate
Waymo’s new system can turn ordinary video into a 3‑D LiDAR‑style point cloud, sparking a flurry of comments about whether cameras alone can ever replace LiDAR.

“The Waymo World Model can convert those kinds of videos…into a multimodal simulation—showing how the Waymo Driver would see that exact scene.” – xnx
“They convert video into a world model representation suitable for 3D exploration and simulation without using LIDAR.” – smallmancontrov

2. Tesla’s camera‑only strategy and safety concerns
Many users argue that Tesla’s reliance on cameras alone is unsafe, especially in adverse weather, and that the company’s safety‑driver model is a crutch.

“I would never put my child in one.” – IhateAI_3
“Tesla’s autopilot model really does feel like what someone who’s never driven a car before thinks it’s like.” – verelo
“Tesla requires safety drivers. I would never trust the FSD on my Model 3 to be even nearly perfect all the time.” – jellojello

3. Remote human assistance (the “Philippines” controversy)
Waymo’s fleet is said to have remote operators who provide high‑level guidance, not direct control, a fact that has been mis‑characterised by some.

“They provide guidance…The Waymo Driver does not rely solely on the inputs it receives…and it is in control of the vehicle at all times.” – Waymo blog (quoted by ddalex)
“They are not remotely driving the vehicles. Waymo asks for guidance in certain situations.” – ddalex

4. Cost, scalability, and the economics of LiDAR
The discussion frequently touches on how LiDAR prices have fallen, the remaining cost of a full sensor suite, and whether the extra hardware is worth the marginal safety benefit.

“Hesai has driven the cost into the $200 to $400 range now.” – hangonhn
“Waymo reduced their LiDAR price from $75,000 to ~$7,500 back in 2017.” – ra7
“Adding LiDAR to every new Tesla seems to be the best way out of this idiotic position.” – jellojello

5. Trust, statistics, and public perception
Users compare crash statistics, question the reliability of data, and debate whether autonomous cars can earn public trust.

“Tesla has more data than Waymo” is some of the lamest cope ever. – schiffern
“Waymo has driven 7.5 B autonomous miles to Waymo’s 0.2 B, but yes Waymo looks like they are ahead.” – smallmancontrov
“I would never trust a camera‑only FSD.” – verelo

6. The broader societal and infrastructure context
The thread also covers how autonomous vehicles fit into public transit, city planning, and the future of driving.

“Self‑driving cars is a dead end technology…better urban planning, etc.” – runarberg
“Automation makes public transit better…automated minibuses that are more flexible and frequent than today’s buses.” – ianburrell
“If there is a robotaxi available to pick up and drop off every passenger…public transit could become subsidized robotaxi rides.” – spaceywilly

These six themes capture the main currents of opinion in the discussion.

🚀 Project Ideas

Video2World: Open‑Source Video‑to‑3D World Model Converter

Summary

  • Converts consumer dashcam or YouTube driving footage into dense, LiDAR‑style point clouds and semantic maps.
  • Enables autonomous‑vehicle companies and researchers to train perception models on billions of hours of real‑world video without expensive sensor suites.

Details

Key Value
Target Audience Autonomous‑vehicle startups, research labs, hobbyists
Core Feature Neural pipeline that ingests monocular video, outputs calibrated 3‑D point clouds, semantic segmentation, and ego‑motion estimates
Tech Stack PyTorch, ROS, OpenCV, CUDA, ONNX export
Difficulty Medium
Monetization Revenue‑ready: $99/month for commercial API access

Notes

  • “If you can generate reliable 3D LiDAR from 2D video, every dashcam on earth becomes training data.” – anupamchugh
  • HN users lament the lack of large‑scale labeled 3‑D data; this tool fills that gap and can spark community‑driven datasets.

SafeStop: In‑Car Emergency Stop System

Summary

  • A hardware‑software stack that instantly pulls a vehicle to a safe spot and locks the doors when the driver or system detects a critical failure.
  • Addresses frustration over hidden “pull‑over” buttons and the risk of stopping in the middle of a highway.

Details

Key Value
Target Audience OEMs, aftermarket retrofit shops, safety‑conscious consumers
Core Feature Redundant brake‑control interface, GPS‑based safe‑zone detection, audible/visual alerts, manual override lock
Tech Stack CAN‑bus interface, STM32 MCU, ESP‑32 Wi‑Fi, Android/iOS companion app
Difficulty Medium
Monetization Revenue‑ready: $199 per unit + $9.99/month for diagnostics

Notes

  • “I would very much like a big red STOP IMMEDIATELY button in these vehicles.” – breckinloggins
  • The system can be sold as an aftermarket kit or integrated into new vehicles, giving HN users a tangible safety improvement.

CityNav: AI‑Driven Narrow‑Street Navigation Toolkit

Summary

  • A dataset and open‑source model suite specifically trained on narrow, two‑way alleys, cobblestones, and historic city centers.
  • Helps autonomous‑vehicle developers overcome the “old‑world” edge cases that current models struggle with.

Details

Key Value
Target Audience Autonomous‑vehicle developers, urban planners
Core Feature 3‑D reconstruction of narrow streets, lane‑free path planning, occlusion‑aware perception
Tech Stack ROS, Open3D, TensorFlow, Mapbox Vector Tiles
Difficulty High
Monetization Hobby (open source) with optional paid mapping data subscription

Notes

  • “Waymo is ramping up in Boston’s cobblestones, narrow alleyways, roundabouts.” – xnx
  • HN commenters like “smallmancontrov” and “tanseydavid” highlight the need for realistic narrow‑street data; CityNav directly addresses that.

RemoteAssist: Low‑Latency Remote Driver‑Assistance Platform

Summary

  • A cloud‑based platform that streams high‑fidelity sensor data to remote operators with sub‑200 ms latency, enabling real‑time guidance in edge cases.
  • Provides a cost‑effective fallback for autonomous fleets without full remote driving.

Details

Key Value
Target Audience Robotaxi operators, fleet managers
Core Feature Edge‑compute video/audio streaming, AI‑assisted decision prompts, operator dashboard
Tech Stack WebRTC, gRPC, Kubernetes, NVIDIA Jetson
Difficulty Medium
Monetization Revenue‑ready: $5k/month per fleet

Notes

  • “Waymo asks for guidance in certain situations and gets an input, but the Waymo vehicle is always in charge.” – Waymo blog
  • HN users such as “turtlesdown11” and “mrcwinn” discuss the need for human backup; RemoteAssist offers a scalable solution.

DepthVision: Adaptive Multi‑Camera Depth Estimation

Summary

  • A hardware‑software stack that mounts a small, low‑cost camera array with adaptive focus and HDR to produce high‑resolution depth maps in all lighting conditions.
  • Eliminates the “camera‑only” perception pain point and reduces reliance on expensive LiDAR.

Details

Key Value
Target Audience OEMs, aftermarket vendors, autonomous‑vehicle hobbyists
Core Feature Real‑time depth from multi‑view stereo + learned depth refinement, dynamic focus adjustment
Tech Stack Raspberry Pi 4, Intel RealSense D435i, OpenCV, PyTorch
Difficulty Medium
Monetization Hobby (open source) with optional paid firmware updates

Notes

  • “Human depth perception uses stereo out to only about 2 or 3 meters… beyond that we use context clues.” – robotresearcher
  • HN commenters like “cobrastanJorji” and “pookeh” emphasize the need for better depth; DepthVision directly tackles this.

LidarCost: Modular Low‑Cost LiDAR Stack for Consumer Vehicles

Summary

  • A fully open‑source, low‑cost LiDAR design (≈$200 per unit) that can be integrated into consumer cars or aftermarket kits.
  • Provides high‑resolution 3‑D sensing without the price barrier of current automotive LiDAR.

Details

Key Value
Target Audience OEMs, aftermarket vendors, autonomous‑vehicle hobbyists
Core Feature 360° coverage, 200 m range, 10 Hz update, open‑source firmware
Tech Stack FPGA (Xilinx Artix‑7), custom laser driver, open‑source CAD
Difficulty High
Monetization Revenue‑ready: $299 per unit + $19.99/month for firmware updates

Notes

  • “Hesai has driven the cost into the $200 to $400 range now.” – bragr
  • HN users such as “mgaunard” and “bragr” discuss LiDAR cost; this project makes LiDAR accessible to a broader market.

Read Later