Advanced Navigation Solutions with Gemini AI: Integrating Augmented Reality for Enhanced Transportation Safety

Ashok Shanmugam¹ , Gajalakshmi Jayaseelan² , Sameeha.SK Basha² , Shaarmi Rajendran² and Prabhu Venkatesan³
1. Associate Professor, Dept of ECE, Vel Tech Multi Tech Dr. Rangarajan Dr. Sakunthala Engineering College Chennai, India
2. UG Scholar, Dept of ECE, Vel Tech Multi Tech Dr. Rangarajan Dr. Sakunthala Engineering College Chennai, India
3. Professor, Dept of ECE, Vel Tech Multi Tech Dr. Rangarajan Dr. Sakunthala Engineering College Chennai, India
Correspondence to: S. Ashok, sashok@veltechmultitech.org

DOI: https://doi.org/10.70389/PJS.100143

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: S. Ashok, J. Gaja Lakshmi, S. K. Sameeha, R. Shaarmi and V. Prabhu – Conceptualization, Writing – original draft, review and editing
• Guarantor: S. Ashok
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: augmented reality vehicle navigation, Gemini ai route optimization, esp32 iot integration, mq-3 alcohol detection, driver distraction audio monitoring.

Peer Review
Received: 14 August 2025
Last revised: 26 September 2025
Accepted: 1 October 2025
Version accepted: 3
Published: 24 December 2025

Plain Language Summary Infographic

Introduction

Gemini AI and augmented reality have the capacity to take navigation a step further and elevate the potential of transportation safety by allowing device to guide humans after analysing data from various sources instantly whereas traditional system have proven to be effective in satisfying our daily travel needs but they do have their share of limitations they rely on static maps lag in updates and compel a driver to look at a screen a potentially dangerous distraction making the response to the next unforeseen event like an accident in front of you or road traffic even more difficult to see the road. Prior system relied on leveraging Augmented reality, Semantic-Segmentation for Enhanced Safety, it aims to explores how augmented reality (AR) can able to connect with semantic segmentation and Vehicular Ad-Hoc Networks (VANETs) to improve driving safety.¹

It ensures real-time hazard warnings, it discusses how this integration improves vehicle-to-vehicle communication and supports real-time hazard detection, offering better situational awareness for drive. This current system works with small low-cost microcontroller called ESP32 which wirelessly connects to a server this server sends real-time navigation updates to the AR glasses so the driver can see the latest directions without having to touch anything.   The proposed work aims to make driving and cycling safer by using augmented reality to show directions data right in front of the driver’s eyes instead of looking down at a phone or GPS device through AR glasses facilitates a hands-free experience. In addition to navigation, the system has an alcohol detection mechanism using an MQ-3 sensor.²

The ignition lock is activated by the system if the rider’s breath alcohol level exceeds the legal threshold, which means the motorcycle will not start. This aspect lowers the danger of drunk driving, increasing safety levels further. A microphone sensor is also there to inform calls so that the rider is not needed to take a look at his mobile for the glasses to notify him, therefore not altering his attention to the road. The next step in road safety improvements is real-time hazard detection in augmented reality navigation.³ These systems identify risks, such as pedestrians, other vehicles or obstacles by joining sensors like cameras with machine learning such systems process the visual data to give quick alerts and warnings on AR displays. Current navigation link failure integration with advanced technologies like augmented reality (AR) and AI which used to supply more clear and deep engaging experience. Most solutions do not include proactive safety features including detection of hazards or driver monitoring.

Current Works

Present vehicle navigation needs to be strong on conventional GPS- based systems. These generally required to split in-car displays or mobile devices that involved drivers to divert their eyes off the road to use route instructions. These kinds of actions lead to distractions from the road and which leads to accidents, so system provide us a fixed or slightly flexible navigation route guidance, failing to account for sudden changes in traffic patterns, road conditions, or unexpected hazards in real-time. This framework has designed to integrate AR with AI-power traffic check to upgrade navigation in live.⁴ Using a camera calibration, the visual processing of the AR prompt will be presented along the central 40-degree horizontal field of view in front of the rider. It used to research how AR filters traffic updates and route adjustments directly onto driver’s field of view. The system maximizes IoT-enabled sensors to get live data and provide proactive alerts for road hazards and congestion, resulting in enhanced safety and efficiency. These gaps in adaptability, user experience, and safety measures call for a more integrated and intelligent solution, like that found in the proposed system combining AR, Gemini AI, and alcohol detection, transforming driving into safe, most of these systems fail to integrate efficiently with real-time data processing abilities, especially in the analysis of traffic patterns and then modification of routes. Most of the navigation applications are based on historical data of traffic or static data about the road, so they miss events that happen like accidents and closure of road.

The dynamic traffic update in AR navigation systems proposed⁵ using AI and IoT to present dynamic guidance. Recent advancements like edge-Al-based navigation systems and federated learning for real-time decision-making in AR frameworks (2023–2024) have improved navigation solutions in terms of low latency and privacy. By analyzing live traffic updates and street blockages. These systems help in avoiding traffic jams and a short wait. Tests reveal 25% improvement in travel time when dynamic updates are applied compared to static GPS routes. Today’s navigation systems are not largely connected with the much larger Vehicle-to-Everything (V2V) ecosystem in which cars talk to other cars as well as to infrastructure like traffic lights and road signs. This prevents the system from respond in real-time to the traffic. Previous studies have shown that an Augmented, Virtual, and Mixed Reality passenger experiences.⁶ This research discusses AR’s role in transforming the in-vehicle experience for passengers, particularly in autonomous vehicles. It explores how AR displays can enhance navigation, entertainment, and interaction while ensuring safety and comfort, making driving or traveling more immersive.

Proposed Methodology

The proposed methodology displays the direction information on the driver’s viewing angle, minimizing interruption from the path. To understand routing data, the direction information is focused on the OLED glass. The conventional steering system wants the driver to look aside from the way this AR journey mapping delivers real-time data and affords effortless navigation, facilitating the riders to keep attentive on the streets. The ESP32 (Bluetooth capabilities)  processor functions as the central unit overseeing Bluetooth interfacing and managing inputs obtained from the Gemini AI platform for direction tracking. The Gemini intelligence server grants live mapping guidance and enhancements that are communicated to the ESP32 device and rendered in the holographic display goggles. It assists cordless interaction among the user’s cellular gadget system, and it increases comfort for handling. In AR glasses, the users are able to keep their hands upon the steering and pay attention towards travelling, which is an advantageous thing for riders. The operators enter directions through wireless technology handled by Gemini AI to identify the best path. It includes intoxication tracking. If alcohol rates identified exceed a defined limit, the vehicle will instantly decelerate and halt.⁷

Figure 1 has two sections: one is AR section and the other is vehicle section. In the AR section, the input is given through the AMR voice app by connecting the HC-05 (Bluetooth) module to that app. Once the data is recognized by the module, it will send the data to ESP32 controller which is connected to Gemini server through the API key. Gemini will provide the optimal navigation route and send it back to the ESP32 and this data will be transmitted to the AR glass wirelessly by using UART protocol which supports serial communication.⁸ The data is shared bit by bit.

Figure 2 portraits the block diagram of vehicle section. ESP8266 is used to share data from the AR section to Vehicle section. The MIC sensor and MQ-3 sensor is fixed near by the operator mouth. The sensors collect the voice frequency and detects ethanol in the air of the user and continuously sends the data to the motor controller. Once a suspicious activity is detected above the threshold level, the ESP8266 of helmet section sends the data to the ESP8266 of vehicle and automatically the motor of the vehicle will be slow down. It reads, Fleet management in logistics is supported through multi-user integration with cloud-syncing in Qt framework. Thus, ESP8266 forms a fundamental role in the navigation system within the AR using ESP32 since it enables a wireless communication that transmits commands to the unit used in viewing AR.

Module Description

Using Bluetooth, rider send routing requests that are passed to an AI module for computation. Once the AI module has computed the best way to respond back to ESP32. Install libraries for HTTP requests and JSON parsing, obtain an API key from AI, send an HTTP query to the AI, and it analyzes the data and sends it back to the processor. To implement the Gemini AI-powered enhanced-navigation system, start with the ESP32 setup.⁹ Install the necessary libraries, which include ESPA sync Webserver for handling HTTP requests and Arduino JSON parsing. State the navigation question explicitly and generate a JSON object that includes the question and the API key. It calls the Gemini API endpoint with an HTTP POST request by correctly formatting the prompt, which is then authenticated by means of the API key.

The ESP32 module draws up to 160 mA (in the active mode), and the OLED display takes around 25 mA, resulting in around 4 hours of run time with a 1000 mAh battery. From the obtained output of this response, further parsing is performed for relevant information that can be displayed on an OLED or maybe subjected to further processing, such as conversion to audio. The cost function f(x) = w1-traffic + w2-hazard + w3-distance, where w1, w2, and w3 are assigned by Gemini Al inference, is used to select routes. Use of robust error handling: This means checking for network problems or an invalid response and including the retry mechanism where requests fail for reliable system performance.¹⁰

Figure 3 represents the operational principle about an ESP32-based system interacting with an API Gemini AI. The ESP32 microcontroller would be initialized in the system by setting up Bluetooth prompts incoming. This is a system that implements Gemini AI. This system uses sensor datasets in order to achieve a low latency inference time (~60 ms) for real-time augmented reality feedback. An HTTP request is formulated based on proper URL and header conditions. The request is then sent to API, and it waits for the response; it is parsed to extract relevant data, which is then transmitted via AR TX. It sends the parsed data wirelessly to AR RX, which the displays the content in AR glass. The system loop back, always waiting for new Bluetooth prompts, always available for continuous interaction with the Gemini AI API and AR glass display.¹¹

Results and Discussion

The MQ-3 sensor measures the concentration of alcohol in the breath sample. If the measurement is below the predetermined threshold, the vehicle portion permits the bike to start. If the measurement level is higher than the threshold, the bike cannot start, and the vehicle is locked. By limiting the number of intoxicated drivers on the road, this will ensure that only sober riders are allowed to drive the vehicle, thereby ensuring road safety and minimizing drunk driving.¹² Table 1 describes the analog output values of the MQ-3 sensor for different events. The minimum value for ethanol detection is 200 and the maximum value is 1023. For normal clean air, the analog output varies between 0 and 200, such that no alcohol will be detected. At 300 ms or more, the network latency rate went up by 18%, pointing at the sensitivity of the connectivity. On being exposed to alcohol fumes, the output from the sensor is substantially high, from 500 up to 1023, while crossing this level, then only the system will slow down.

Table 1: Analog output values of MQ-3 sensor.
Event	Minimum Value (ADC Counts)	Maximum Value (ADC Counts)
Ethanol detection	200 ADC	1023 ADC
Normal clean air	0 to 100 ADC	200 ADC
Exposure to alcohol fumes	500 to 800 ADC	1023 ADC

Such information manifests the ability of this system in providing discrimination between regular air and one carrying alcohol, which results in perfect alcohol detection coupled with high security measures that will effectively distinguish normal air from air full of alcohol, which would result in good detection of alcohol and increased safety alcohol fumes, the output from the sensor is substantially high, from 500 up to 1023, while crossing this level, then only the system will slow down. Such information manifests the ability of this system in providing discrimination between regular air and one carrying alcohol, which results in perfect alcohol detection coupled with high security measures that will effectively distinguish normal air from air full of alcohol, which would result in good detection of alcohol and increased safety.          

Figure 4 shows the output graph of the MQ-3 sensor clearly exhibits how well this sensor functions during the times is available. The X-axis represents different air conditions, and the Y-axis represents the analog output in ADC values (0–1023) In this case, there is a high spike in the analog output values for alcohol fumes, reaching values as high as 1023. The reading remains below 200 in regular air. In this graph, it was observed that the sensor is extremely sensitive and responded in real time, which gives it a reason to be regarded as a useful component in intoxicated drivers’ detection to enhance vehicle safety systems.

The integration of MIC sensor detection with vehicle control, is necessary for ensuring the success and safety of this technology in vehicle control. Consultation would assist the advice in the best practices to implement this technology, optimize its performance, and prevent possible challenges. AI-based navigation systems will help in understanding how real-time data processing, signal recognition, and response mechanisms occur.¹³ False-positive rate of the vehicle signal detection propensity within the system was observed to be as high as approximately 12 percent. The cause of the same was identified to be non-acoustic signatures being emitted by vehicles, such as the honking of car horns or the sound of accelerating engines. The calibration settings for the MIC sensor, appropriate hardware components to be selected, and the algorithmic approach for filtering out non-call sounds to avoid false positives would also require consultation. In addition, consultations can guide the development of user-friendly interfaces for drivers, so that the system doesn’t distract itself. Collaboration with experts can further assist in matters of regulatory compliance and safety standards, ensuring the traffic and safety regulations are lawful.

In normal ambient sound conditions, the output ranges from 0 to 200, with a quick response time of 100 milliseconds. For low noise, the outputs increase to 100–300 with a 200 ms response. High noise and acute sounds give outputs between 400–1023 and 600–1023, with response times between 300ms and 150ms respectively. This data depicts the fact that the sensor is accurate in detection of sound variation and responsive in terms of scenarios. The threshold for the MQ-3 sensor was empirically calibrated at nearly 300 ADC counts, which corresponds to 0.4 V, in order to segregate normal and intoxicated states. The MIC sensor was calibrated to perform within ~500 ADC counts for the activation when it senses a sharp sound. Table 2 defines the efficiency of the MIC sensor is reflected by the graph plotting its response time against different types of events such as normal sounds, low noise, high noise and sharp sounds (e.g., Clap)-triggers detection of incoming calls. It responds to normal sounds in 100 ms and sharp sounds in 150 ms.

Table 2: Analog output values of MIC sensor.
Event	Minimum Value (ADC Counts)	Maximum Value (ADC Counts)
Normal Ambient Sound	0 ADC	200 ADC
Low Noise Detection	100 ADC	300 ADC
High Noise Detection	400 ADC	1023 ADC
Sharp Sound (e.g., Clap)	600 ADC	1023 ADC

Figure 5 indicates the role of the sensor in the detection of calls and distraction for a safer driving experience with accuracy and at the right time. The X-axis represents different sound events, and the Y-axis represents the analog output in ADC values (0–1023) The software output of the system is the core element driving the functionalities of the AR navigation glasses. The output is generated through the collaboration of the ESP32 microcontroller, Bluetooth module, Gemini AI server, and the AR projection interface. Once the user inputs a destination through Bluetooth, the ESP32 microcontroller processes this data and establishes a connection with the Gemini AI server via Wi-Fi.¹⁴ The system’s average real-time processing time from event detection to presentation of alerts was computed to be 250 ms. WPA2 is the preferred Wi-Fi communication security of the ESP32-end devices; later versions of which would introduce end-to-end encryption very soon to eliminate chances of data breaches,’ it said.

Figure 6 displays the software output of the system which is the core driver functionalities of the AR navigation glasses. The output is produced by the combination of the ESP32 microcontroller, Bluetooth module, Gemini AI server and the AR projection interface. After a user input a destination through Bluetooth, the ESP32 process this information and connects with AI through Wi-Fi. The system utilizes a 0.69-inch OLED display (model: SSD1306,128×64 resolution) with 12C interface to present navigation alerts and sensor output to the rider in real time. The server computes the best route according to real time navigation.

The system demonstrated in the above figure implements a navigation system that involves backend processing through text parsing, filtering and producing structured output. The Arduino IDE environment is used for execution. The main parts of this system are the microcontroller platform for processing, serial communication with regard to data transmission and custom algorithms for input-output manipulation. The code accepts a raw string query, an example of which is a request for a navigation route from Chennai Tambaram to Chennai central. This query will be passed through the filtering algorithm. In example code, for loop iterates over the input string to delete unwanted characters in conditions such as isalnum () to retain alphabets or numbers and is space () that allows spaces to pass through. It eliminates all unwanted symbols or noise in the input data so that it can become a clean, usable string. This parsed query is then transferred to a parsing function, which would most probably be in the form of defined rules or a lookup table in which the text could be converted to actionable navigation instructions. Outputs occur in a serial format such as “turn right”, “go straight”, and turn left.

These serial outputs occur always in real time on the serial. println (), which is basic to the serial communication library associated with the Arduino framework. The system implements a deterministic way of timing and ordering its outputs. Each step in the navigation must have incorporated a buffering process that ensures that data displayed is in an orderly sequence. Serial communication parameters, such as the baud rate,115200, were optimized for the highest speed for data transmission; hence, easy communication between the microcontroller and the IDE would be ensured. Although here it is represented in a simulated environment, in principle, such a system easily could be used with GPS modules map API’s or some other navigation detected hardware the response time and accuracy in delivery instructions are dictated by the computational efficiency of filtering and algorithm.¹⁵

The serial monitors display data indicating a robust back-end system that has successfully completed navigation request handling it demonstrates how a full up real time embedded system uses it to achieve task around it, processing text in out as well as filtering characters and managing serial output modular and efficient programming principles ensures scalability and integration capabilities for more complex navigation solutions. An AR picture noticed they can offset the normal percentage of the turn missed when objects were not at cost and it was inclusive of 90 percent real rider negligence and commitment.                    

Figure 7 shows the experimental setup of the proposed AR-based navigation and safety system using Vufine+ AR glass and SSD1306 OLED display. ESP32 microcontroller is a powerful central processor coordinating communication back to the input device and even to the AR display. The research conducted a series of tests in urban locations subjected to tight traffic and semi-rural conditions during varied situations of light and sound. The destination would be input by the rider via Bluetooth using ESP32, which would further process this information into data to transmit over Wi-Fi to the Gemini AI server. Based on the input, the destination provided, the Gemini AI server decides upon the best route to follow. This then is sent to the ESP32 in order to compute and send for display to the AR glasses.  It is overlaid on the rider’s direct field of view as directional arrows or distance markers for seamless and intuitive navigation.

Figure 8 integrates the detection of mobile phone calls with MIC sensor closely in line with the system objectives. It means no user-identifiable data is saved or transmitted, so the rider’s privacy is kept safe. Consultation sessions provided actionable insights on best practices on maximizing system performance and addressing possible challenges. The enhanced navigation system was shown to be a viable option that successfully demonstrated its functionality and impact on convenience and safety.                                       

Conclusion and Future Works

This motorcyclist safety system integrates augumented reality (AR) to provide real time navigation data instruction to enhance the transportation safety. It knows forgiving it a devout performance, it hinges on link-up stability in a controlled audio environment. This system utilizes the ESP32 for communication and control purposes of transmitting data seamlessly between Gemini AI, AR glasses and sensors via serial communication protocols. Future work includes bringing the LiDAR sensor to avoid obstacles and using it for autonomous e-scooter handling tasks. The data will be display in the text format which is an accurate navigation data and it offers hand free operation for the riders.

Adding the MQ-3 alcohol detector sensor and microphone sensor, it facilitates further safety with the capability of controlling motor speed. In conclusion, The upcoming developments involve adding voice command interfaces for hands-free operation, more real-time processing of data via edge AI features, and enhanced compatibility with LiDAR modules for obstacle avoidance and this system attained its intended objectives. It provides a functional, safe and efficient navigation aid for motorcyclist. It gives a base that can be developed further to add more sensors, expand the means of inputting data, or even use of IoT connectivity for a broader application. The prototype complies with standard and operates within safety regulations for local conditions concerning individual transport devices.                                

References

1. Afzal S. et al., “Leveraging AR, semantic segmentation, and VANETs for improved driving safety,” IEEE Transactions on Vehicular Networking, 2024; 22(4): 389-400.
2. Ameen H. et al., “Sensor-based alcohol detection system for vehicle safety using MQ-3 sensor,” IEEE Transactions on Vehicular Safety Systems, 2023; 29(4): 1119-1132.
3. Chen R. et al., “Real-time hazard detection in AR navigation using LiDAR, cameras, and machine learning for improved road safety,” IEEE Journal of Augmented Reality, 2023; 31(4): 215-227.
4. Evans L. et al., “Integrating AR with AI for live traffic updates and enhanced driver safety in connected vehicles,” IEEE Transactions on Connected Vehicles, 2024; 19(1): 65-76.
5. Jones M. et al., “Dynamic traffic updates in AR navigation systems using AI and IoT for route optimization and smoother traffic flow,” IEEE Transactions on Transportation Systems, 2024; 16(3): 254-265.
6. V. Nehru and V. Prabhu (2023) Segmentation of Brain tumor sub regions with Depth wise Separable Dense U-NET (DSDU-NET), International Journal of Imaging Systems and Technology., 2023; 33(4): 1323-1334, 2023 https://doi.org/10.1002/ima.22861.
7. Miller D. et al., “AR and IoT integration for accident prevention and real-time hazard detection,” IEEE Internet of Things Journal, 2023; 25(5): 450-460.
8. Nguyen P. et al., “Holographic displays in AR navigation systems for 3D route guidance and touchless controls,” IEEE Transactions on Holographic Imaging, 2024; 14(2): 158-170.
9. Patel S. et al., “Connected vehicle communication for AR navigation and real-time road hazard alerts using Vehicle-to-everything V2X,” IEEE Transactions on Vehicular Technology, 2023; 72(6): 1890-1902.
10. Pradeepkumar G. et al., “Safe transportation system with IoT-based alcohol detection and automated vehicle immobilization,” IEEE Journal of Safety and Security Engineering, 2024; 6(8): 505-517.
11. Raja D. P. et al., “IoT-based alcohol detection system for drunk driving prevention and real-time emergency alerts,” IEEE Transactions on Intelligent Transportation Systems, 2023; 27(9): 1502-1513.
12. Reddy K. et al., “Smart helmet integration with AR for enhanced safety in hazardous environments,” IEEE Sensors Journal, 2024; 23(7): 1927-1939.
13. Sarkkinen T. et al., “Rock-solid AR: A platform for automotive AR development with AI-driven localization and motion tracking,” IEEE Journal of Automotive Systems, 2024; 18(2): 204-218.
14. S Sivasaravanababu, V.Prabhu, V Parthasarathy, Rakesh Kumar Mahendran (2021),An efficient epileptic seizure detection based on tunable Q-wavelet transform and DCVAE-stacked Bi-LSTM model using electroencephalogram, The European Physical Journal Special Topics, https://doi.org/10.1140/epjs/s11734-021-00380-x
15. Wang D. et al., “AI-driven navigation systems for enhanced driver-road connection through real-time metrics and advanced algorithms,” IEEE Transactions on Intelligent Transportation Systems, 2023; 28: 1234-1245.

Cite this article as:
Ashok S,  Gaja Lakshmi J,  Sameeha SK, Shaarmi R and  Prabhu V. Advanced Navigation Solutions with Gemini AI: Integrating Augmented Reality for Enhanced Transportation Safety. Premier Journal of Science 2025;15:100143

Export to EndNote Export to Mendeley