Autonomous Mobile Robot for Indoor Navigation and Real-Time Obstacle Avoidance using SLAM and Edge AI Accelerators

Project Abstract

This research presents the design, implementation, and evaluation of an autonomous mobile robot capable of indoor navigation and real-time obstacle avoidance by leveraging simultaneous localization and mapping (SLAM) and edge AI accelerators for on-board perception, planning, and control. The core objective is to achieve robust autonomous operation in cluttered indoor environments with limited GPS and variable lighting, while ensuring real-time performance, energy efficiency, and safety. The system integrates a lightweight LiDAR-based SLAM (RAM-SLAM) with visual-inertial odometry to produce accurate maps and pose estimates, enhanced by loop closure and multi-sensor fusion to reduce drift over extended missions. A modular perception pipeline combines LiDAR, RGB-D, and event-based cameras to detect static and dynamic obstacles, classify objects, and estimate motion trajectories of moving agents. Real-time obstacle avoidance and path planning are driven by a hybrid planner that merges global A* or Dijkstra-based routes with local dynamic window and model predictive control (MPC) for smooth, collision-free trajectories in dynamic environments. To accelerate on-board computations and reduce latency, the robot employs edge AI accelerators (such as embedded GPUs, TPU-like accelerators, and neural processing units) to run lightweight deep learning models for obstacle classification, semantic mapping, and motion prediction, while maintaining low power consumption suitable for long-duration operation. The research also investigates adaptive sensing strategies and data fusion schemes to cope with sensor noise, occlusions, and varying power budgets, including fault-tolerant behavior and recovery mechanisms in the face of sensor failure. A comprehensive experimental methodology is presented, including simulation-based validation, bench-top experiments, and real-world indoor tests across multiple layouts (corridors, rooms, and cluttered areas) with varying levels of dynamic activity. Key performance metrics include localization accuracy, map quality (completeness and consistency), obstacle detection range and latency, planning success rate, path efficiency, energy consumption, and safety assurances under fault conditions. Results indicate that the integrated SLAM framework achieves centimeter-level drift over tens of meters with robust loop closure, while the perception stack maintains high recall and precision for dynamic obstacles under challenging lighting and occlusion. The edge AI accelerators deliver substantial speedups for perception and motion prediction tasks, enabling end-to-end closed-loop control at frame rates suitable for real-time operation, with measurable improvements in latency and energy efficiency compared to CPU-only baselines. The findings demonstrate feasibility of deploying a modular, scalable autonomous robot capable of reliable indoor navigation and proactive obstacle avoidance in real-world settings, along with insights into the trade-offs between sensing modalities, processing architectures, and control strategies. The work also outlines design guidelines for deploying similar edge-enabled autonomous systems, including dataset collection protocols, evaluation benchmarks, and open-source software components to foster reproducibility and further research in indoor robotics and edge AI-enabled perception.

Project Overview

What This Project Is About

---

A hands-on study of a small robot that can move around indoors, find its way, and avoid obstacles in real time. It uses sensors to map its surroundings and simple computer routines to decide where to go next, with extra power from specialized hardware that runs these tasks quickly.

---

---

The Problem It Addresses

---

Many mobile robots struggle in unfamiliar indoor spaces where clutter, dynamic people, and changes in lighting can cause misdirection or collisions. This project aims to make navigation safer and more reliable without needing powerful computers, by using efficient sensing, mapping, and local decision-making.

---

---

Objectives of the Project

---

1. Build a small indoor robot platform with position-tracking sensors and a camera or lidar for sensing obstacles.
2. Implement a lightweight mapping method to understand the robot’s surroundings.
3. Enable real-time obstacle avoidance to prevent collisions.
4. Incorporate edge AI accelerators to run core tasks efficiently on-device.
5. Test navigation in different indoor layouts and measure performance improvements.

---

---

What You Will Do Step by Step

---

1. Gather hardware: robot chassis, sensors, and a compact edge AI processor. 2. Learn basic sensing and control basics (movement, reading sensors). 3. Implement a simple map-building routine and obstacle detector. 4. Add real-time decision rules to navigate without collisions. 5. Optimize code to run on the edge device. 6. Create and run test scenarios, collect data, and compare results. 7. Document findings and reflect on limitations.

---

---

Expected Outcome

A functional indoor robot capable of mapping its environment, avoiding obstacles in real time, and running efficiently on an edge AI device. The project should show safer navigation with quicker on-device processing and provide insights for future improvements in similar, low-power robotic systems.