Kotlin Tutorial Shows How to Build a Voice-Controlled Robot Using LLMs

·1 views

A developer tutorial published on DEV Community outlines how to build an Android app that lets users control a robot through spoken natural-language commands. The system uses Android's speech-recognition API to convert voice input into text, which is then parsed by a large language model into structured robot commands. A key design principle emphasized is that the LLM must never directly control the robot; instead, its output must pass through a deterministic safety validation layer before execution. The validated commands are then sent to a robot gateway that translates them into ROS 2 instructions, while the Android app displays live robot status via a Compose dashboard. The tutorial also covers handling ambiguous commands, combining voice with vision for tasks like person-following, and choosing between cloud-based or edge AI deployment depending on latency and privacy needs.

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)

Password Hashing Speed Varies Significantly Across CPU Architectures and Languages

A developer tested PBKDF2 password hashing performance across different hardware and programming languages, finding notable differences between Apple M4 Pro and AMD Ryzen 9 9950X processors. On the same task with 500,000 iterations, the Apple M4 Pro consistently outperformed the AMD Ryzen chip across Rust, Odin, and Zig implementations. Even within a single language like Rust, the choice of cryptographic library — aws-lc-rs versus ring — produced measurable performance differences. The findings highlight that code developed on modern Apple Silicon hardware may behave differently when deployed to x86-64 cloud servers. Developers are advised to benchmark security-sensitive operations directly on production hardware to avoid unexpected performance bottlenecks.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why 'It Works on My Machine' Is a Dangerous Assumption for Developers

The phrase 'it works on my machine' reflects a common but risky gap between development and production environments, stemming from differences in operating systems, browsers, screen sizes, and hardware. A developer discovered this firsthand while updating a deprecated password-hashing algorithm that performed well on their laptop but ran 10 times slower in the cloud test environment. Investigation revealed the test servers were over a decade old, kept running for cost reasons, while even production servers showed a 2x performance drop compared to the developer machine. Factors like Apple's M-series chips powering modern developer laptops versus older x86-64 cloud servers can create significant performance mismatches. The experience underscores that code must be evaluated in the environment where it will actually run, not just where it was written.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

OmniMem extension monitors IDE RAM usage across VS Code and its forks

A developer has released OmniMem, an open-source VS Code extension that displays real-time RAM consumption directly in the editor's status bar. The tool was built after existing memory-monitoring extensions stopped working when the developer switched from VS Code to forks like Windsurf, Cursor, and Devin Desktop. Unlike older solutions that relied on hardcoded process names, OmniMem uses runtime information to identify the active editor and its related processes, ensuring broader compatibility. The extension supports multiple display formats — including used memory, total memory, and percentage — with a configurable refresh interval. OmniMem is available on both the Visual Studio Marketplace and Open VSX, making it installable across a wide range of VS Code-based editors.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Bluesky Embeds Its Logo Invisibly Into User Screenshots

Bluesky, the decentralized social media platform, has been found to embed its logo into screenshots taken by users. The discovery was shared online and drew attention from the tech community on Hacker News. The technique involves rendering the logo as part of the screenshot output, even when it is not visibly prominent in the original interface. This practice raises questions about transparency, as users may not be aware their screenshots are being watermarked. The finding has sparked discussion about platform control over user-generated content and image metadata.

0 comments Read more at Hacker News