The Dynamic and Autonomous Robotic Systems (DARoS) Laboratory at the University of Massachusetts Amherst in the Manning College of Information and Computer Sciences is creating robotic systems for common goods. Our primary research area is in the dynamic locomotion of legged systems with a focus on the development of control architectures and their experimental validation. The ultimate goal of our group is to develop robots to be practical tools for human life by enhancing robotic systems to be faster, smarter, and more robust.
Project Description
The overarching goal of our research is to develop a quadruped robot that can assist with navigation for people with visual impairments. For this ERSP project, we will focus on developing a neural network to identify objects that are critical for blind individuals (e.g., tactile paving, crosswalk buttons for accessibility, stair handrails, and elevator braille), which are currently beyond the capabilities of existing foundation models.
The main activities will include collecting training data. One interesting aspect is that these images must be captured from the perspective of a robot dog. Therefore, the student will learn how to operate the robot, save image data, and perform post-processing.
Another method for collecting data is through simulated environments. Realistic data can be obtained from simulations such as Unreal Engine or IsaacSim. Studying how effectively simulation data can be used compared to real-world data will be an interesting research question. Some coding will likely be required for this task.
Learning Objectives:
Know how to operate a quadruped robot, gather image data, post-processing, synthetic data generation in IsaacSim, (potentially) neural network training
Skills to learn:
Python coding.
General understanding about quadruped robots (watch robot dog YouTube videos).