AI Decodes Japanese AC Remote

It was 4 AM in a Tokyo hotel room. Michael S. Galpert, an American traveler, woke up drenched in sweat. The room was unbearably hot, but when he reached for the air conditioner remote control, he encountered an unexpected obstacle: every single button was labeled in Japanese characters. In that sleepless moment, with no knowledge of the language and no front desk to call, he faced what millions of international travelers have experienced the frustration of being unable to operate basic appliances in a foreign country.

Instead of pressing buttons randomly and hoping for the best, Galpert pulled out his smartphone and turned to artificial intelligence. Using Google Gemini Nano Banana, an experimental AI image generation and translation tool, he snapped a photo of the remote and prompted the AI with a simple request: “Create an image in English”. Within seconds, the tool identified every button on the remote, translating labels for temperature control, fan speed, operating modes, timer settings, and more. He adjusted the temperature and went back to sleep.

Galpert later shared his experience on X (formerly Twitter), posting before-and-after images of the remote. His post quickly went viral, amassing thousands of shares and comments from travelers who had faced similar struggles abroad. “The Nano Banana gods saved the day,” he wrote. The story resonated because it highlighted something universal: the helpless feeling when language becomes a barrier to basic comfort, and the growing power of AI to dismantle those barriers in real time.

But beneath this simple, relatable anecdote lies a fascinating intersection of technologies. How does AI actually decode a Japanese AC remote? What makes air conditioner remote controls so notoriously difficult to understand and replicate? And what does this mean for the future of travel, smart homes, and human-computer interaction?

This article explores the technology behind AI-powered remote decoding, the unique challenges posed by Japanese air conditioner remotes, the broader implications for travelers and smart home enthusiasts, and the future of AI in everyday problem-solving.

Part 1: Understanding the Challenge – Why Japanese AC Remotes Are So Difficult

A. The Language Barrier

Japanese air conditioner remotes are notorious among international travelers. Unlike many consumer electronics that feature English labels or universally recognized icons, Japanese AC remotes often use kanji, hiragana, and katakana characters exclusively. Buttons labeled “冷房” (cooling), “暖房” (heating), “除湿” (dehumidification), “風量” (fan speed), and “タイマー” (timer) are incomprehensible to non-Japanese speakers. Even travelers who have some knowledge of Japanese may struggle with the specific terminology used in appliance controls.

B. The Complexity of Air Conditioner Infrared Protocols

Beyond the language barrier, air conditioner remotes are technically complex. Unlike television remotes that typically send a simple, fixed command for each button press, air conditioner remotes operate differently. When you press a button on an AC remote, it often transmits the entire device state—including current temperature setting, operating mode, fan speed, swing position, and timer settings—rather than a single command. This is known as a “stateful” remote control protocol.

For example, when you press the temperature increase button, the remote doesn’t just send a “temperature up” command. Instead, it calculates the new target temperature, updates its internal state, and transmits a complete data packet containing all current settings. This means that the infrared signal from an AC remote can be hundreds of bits long, encoding a snapshot of the entire device configuration.

C. Proprietary Encoding Schemes

Each air conditioner manufacturer Daikin, Panasonic, Hitachi, Toshiba, Mitsubishi, Fujitsu, Sharp, and others—uses proprietary encoding schemes for their infrared signals. These schemes vary in:

Carrier frequency: Most use 38 kHz modulation, but some use different frequencies.
Pulse timing: The duration of marks (infrared light on) and spaces (infrared light off) varies between manufacturers.
Data encoding: Some use pulse position modulation (PPM), others use pulse width modulation (PWM), and still others use more complex encoding methods.
Frame structure: The number of bits, presence of headers, checksums, and repeat frames all differ.

This fragmentation means that a universal remote control must either contain an extensive database of known codes or possess the ability to learn and decode new protocols on the fly.

D. The Specific Case of Japanese AC Remotes

Japanese AC remotes add another layer of complexity. Many Japanese manufacturers implement unique protocols that are not well-documented outside of Japan. For instance, the Hisense HA-S22CE8-W air conditioner uses a frame structure that includes 184 bits (23 bytes) transmitted as a single subframe. The remote doesn’t send a separate main frame; instead, the main control data is embedded within the subframe’s final bytes. Fan speed settings are distinguished by specific byte values, and the entire frame includes a checksum calculated using XOR operations on multiple bytes.

Part 2: How AI Decodes Infrared Signals

A. The Traditional Approach: Database Matching

Traditionally, decoding an unknown remote control involved one of two approaches:

Code Database Matching: The device compares the captured infrared signal against a pre-existing database of known remote control codes. If a match is found, the device can identify the button pressed and the corresponding action. This is how most universal remotes and smart IR blasters work.
Manual Reverse Engineering: For unknown protocols, engineers would capture raw timing data from the infrared signal and manually analyze the pulse patterns to deduce the encoding scheme. This is a time-consuming process that requires specialized knowledge of infrared protocols and signal analysis.

B. The AI Revolution in Remote Decoding

Artificial intelligence has introduced new paradigms for decoding remote controls:

1. Computer Vision and OCR for Interface Translation

The simplest application of AI—and the one used by Galpert—involves computer vision and optical character recognition (OCR). The AI tool analyzes an image of the remote, identifies text labels on buttons, and translates them into the user’s preferred language. This approach doesn’t decode the infrared signal itself; it translates the user interface. However, it solves the immediate problem for travelers: understanding what each button does.

Google Gemini Nano Banana (formally known as Gemini 2.5 Flash Image) combines image recognition with real-time translation. It’s designed for creating and editing images, but travelers have discovered it works beautifully for decoding foreign packaging, street signs, and household appliances.

2. Machine Learning for Protocol Identification

More advanced AI applications use machine learning to identify infrared protocols automatically. By training neural networks on thousands of infrared signal samples from different manufacturers, AI systems can:

Recognize the carrier frequency and modulation scheme
Identify the protocol family (NEC, Sony SIRC, Philips RC-5, etc.)
Parse the data structure and extract meaningful commands
Map buttons to their functions even without a pre-existing database

3. Pattern Recognition for Unknown Protocols

When an infrared signal doesn’t match any known protocol, AI can still provide useful information. A decoder can calculate a stable fingerprint from normalized durations, label it as an unknown protocol, and store multiple captures of the same button for comparison. If repeated captures cluster around the same pattern, the application can map that fingerprint to an action even before the protocol is formally decoded.

4. Large Language Models for Natural Language Control

Some systems now integrate large language models (LLMs) to interpret natural language commands and translate them into device control actions. A user might say “make it cooler” or “set the temperature to 24 degrees,” and the LLM interprets the intent, maps it to the appropriate device command, and triggers the infrared transmission.

Part 3: The Technical Deep Dive – How Infrared Remote Controls Work

To fully appreciate what AI is decoding, it’s essential to understand the underlying technology of infrared remote controls.

A. Infrared Signal Basics

Infrared remote controls communicate with devices by emitting pulses of infrared light. These pulses are modulated at a specific carrier frequency—typically 38 kHz for consumer electronics. The receiver in the appliance detects these pulses and demodulates them to extract the underlying data.

B. Pulse Timing and Encoding

The data is encoded in the timing of the pulses. Two common encoding methods are:

Pulse Width Modulation (PWM): The duration of the pulse (mark) or the gap between pulses (space) encodes binary data. For example, a short space might represent a “0” and a long space might represent a “1”.
Pulse Position Modulation (PPM): The position of the pulse within a fixed time window encodes the data.

C. Frame Structure

A typical infrared frame includes:

Header: A distinctive pulse pattern that signals the start of a frame.
Data Bits: The actual command data, which may include address bits, command bits, and status information.
Checksum: Error-checking data to ensure the frame was received correctly.
Trailer: A final pulse to mark the end of the frame.

D. Stateful vs. Stateless Remotes

Stateless Remotes (typical for TVs): Each button press sends a simple, fixed command. The device maintains its own state independently.
Stateful Remotes (typical for ACs): Each button press sends the complete device state. This allows the remote to display current settings on its screen and ensures that the device and remote stay synchronized.

E. Repeat Frames

When a button is held down, remotes handle repeats differently:

Some resend the full frame repeatedly.
Others send a short repeat frame after the first full command.
The decoder must handle both scenarios correctly.

Part 4: Real-World Applications and Products

A. Smart IR Blasters with AI Learning

Several products now incorporate AI-powered learning capabilities:

i-Ctrl AC Plus: This smart remote controller by AIFA includes an extensive infrared code library covering 95% of infrared appliances and air conditioner brands, including Hitachi, Panasonic, Daikin, and others. It supports app remote control, voice commands through Google Home, Alexa, and Siri, and even includes AI-based mold prevention features.
Surco: A cloud-based smart remote that can control TVs, lights, air conditioners, DVD players, speakers, robot vacuums, fans, set-top boxes, and air purifiers simultaneously.
eRemote: Products like the LinkJapan eRemote mini offer compact, high-performance smart remote control with AI learning capabilities, supporting both Alexa and Google Home integration.