VIDRAFT Adds Korean Voice Control to Boston Dynamics Spot Without Altering Firmware

·4 views

South Korean AI startup VIDRAFT deployed an on-device Korean voice recognition system for Boston Dynamics' Spot robot at the Seoul Robot & AI Museum (Seoul RAIM) on July 23, 2026. The solution enables visitors to issue five Korean voice commands — including 'sit,' 'greet,' and 'lie down' — with Spot responding accordingly. Crucially, the integration required no changes to Spot's hardware or manufacturer firmware, instead layering an AI module on top of the robot's existing control interface. All voice processing happens locally on the device, meaning no visitor audio is transmitted to external cloud servers. Seoul RAIM has signaled plans to expand the system beyond its current fixed command set toward open-ended conversational interaction in a future phase.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

SiteBleed Tracks Real-Time Revenue Loss During Website Downtime

A developer built SiteBleed, an uptime and SSL monitoring tool that calculates live revenue loss during website outages. Unlike standard uptime monitors that only report binary up/down status, SiteBleed shows users a running dollar cost based on their monthly revenue input. The tool displays metrics such as cost per minute and total loss accumulated during an outage, aiming to make downtime tangible for both technical and non-technical stakeholders. It also supports public status pages to help businesses communicate with customers during incidents. Built with React and Supabase, SiteBleed is available at sitebleed.com with a free starting tier.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Alibaba's Qoder IDE Offers Deep Discounts and Transparent Pricing for AI Coding

Qoder, an AI-powered coding tool developed by Alibaba, has drawn attention from developers for its aggressive pricing model built around the Qwen 3 model family. The platform offers up to 98% discounts during off-peak hours and a free 'Lite' mode with no apparent usage cap, making it significantly cheaper than comparable Western alternatives. One developer reported processing around 40,000 lines of code for approximately $10 using Qoder's discounted rates, compared to hundreds of dollars spent on Azure and Google cloud credits for similar workloads. Qoder also provides real-time, itemised cost breakdowns, a feature the developer contrasted with the delayed and opaque billing practices of major Western cloud providers. Additional features include 'Better Loops', a codebase analysis tool that identifies issues and generates scoped fix tasks, and an automated wiki generator that doubles as active codebase context.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to fix missing pip on minimal Ubuntu CI agents without sudo or apt

Developers running self-hosted CI/CD agents on minimal Ubuntu images often encounter 'No module named pip' errors that have nothing to do with network connectivity. Debian and Ubuntu deliberately strip pip and ensurepip from the base python3 package, requiring separate apt packages that are unavailable on no-sudo, minimal CI environments. Attempts to work around the issue using ensurepip or Python virtual environments fail for the same underlying reason, since venv's pip-bootstrapping also depends on the missing ensurepip module. The reliable fix is to use get-pip.py, a self-contained bootstrapper maintained by the Python Packaging Authority that installs pip directly without relying on ensurepip, apt, or root access. Before applying any fix, developers should first verify the agent has outbound network access to PyPI, as firewall or egress issues can produce similar symptoms.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Safely Embed JPG Images in Production PDF Pipelines Without Breaking Accessibility

Embedding JPG images into existing production PDF pipelines is more complex than a simple file conversion, as it involves precise placement within layouts and compliance with accessibility and archival standards. Engineers typically encounter three scenarios: inserting images into templated reports, bulk-ingesting scanned attachments, and generating PDF/A-compliant documents for long-term record-keeping. For templated reports, tools like Puppeteer, Playwright, or LaTeX handle image placement reliably, while server-side libraries such as pdfkit or ReportLab are better suited for batch processing. PDF/A compliance adds further complexity, with font embedding being a common failure point when webfonts sourced from CDNs cannot be legally embedded. Each pipeline type carries distinct failure modes, and conflating them can lead to features that pass initial testing but later fail regulatory or accessibility audits.

0 comments Read more at DEV Community