Robotics Thesis Proposal
- Remote Access - Zoom
- Virtual Presentation - ET
- XUNING YANG
- Ph.D. Student
- Robotics Institute
- Carnegie Mellon University
Teleoperation via Intuition: Safe and Intent Oriented Navigation
This thesis aims to enable seamless teleoperation of a mobile robot by a human operator, such that the robot navigates in unstructured environments following the operator’s intent intuitively, safely, and efficiently. The roles of the human and robot are disproportionate in traditional teleoperation: The human is responsible for most of the autonomy of the robot, including planning, and safety, and dynamic feasibility, and provides an RC input to the robot; whereas the robot is simply responsible for following the provided input. We propose trajectory-based teleoperation, which shifts some of the responsibilities from the operator to the robot by increasing the autonomy of the robot with generation of dynamically feasible and safe local trajectories. Further, we propose modeling and predicting long-term intended paths of the operator, such that trajectories can be generated to complete the task without requiring the operator to explicitly specifying their intention. By increasing the intelligence of the robot, the system requires minimal effort from the operator in order to complete the task.
To this end, we first address the thesis problem by moving mobile robot teleoperation from providing control-level inputs to providing state-space motions via motion primitives based teleoperation with reactive collision avoidance. We then extend the planning horizon with active obstacle avoidance by introducing local trajectory based teleoperation. Instead of reactively pushing the vehicle away from nearby obstacles, we actively generate trajectories that allows the vehicle to follow its intended motion while circumventing local horizon obstacles using a Motion Primitive Tree generated using Biased Incremental Action Sampling (BIAS). Lastly, we build upon local trajectory generation and introduce hierarchical teleoperation, which includes longer horizon intention as global plans.
The proposed research aims to expand trajectory-based teleoperation to include explicit modeling of the operator's intention via predictive teleoperation. We propose to capture the operator's intention in teleoperated navigation tasks as a global path, and reason about the operator's decision making process. The goal of the proposed research is to reason about the operator's path choices in two distinct scenarios: First, in constrained environments where the set of possible paths are severely limited to a set of discrete choices; and second, in open spaces where the operator's intended path could span a continuum of possible paths.
Nathan Michael (Chair)
Helen Olenyikova (Microsoft Mixed Reality and AI)
Sanjiban Choudhury (Aurora Innovation)
Zoom Particpation. See announcement.