Research Focus

Advancement in hardware capabilities, computation power, and control algorithms have physically enabled robots to perform highly complex tasks. Representative examples can be laundry folding, composite sheet layup, liquid pouring, autonomous navigation, mobile manipulation, surgery, etc. Many applications require high degrees of freedom systems where multiple robots can collaborate to perform tasks that are not physically possible by a single robot. Certain applications also require robots to quickly and sufficiently estimate task performance models that are not known a priori. However, it takes significant amount of time, effort, and cost to manually program robots to perform such sophisticated tasks. We could use robots in complex service, manufacturing, and disaster rescue applications, where the task description and the environment is changing frequently, if robots could program themselves.

At RROS, our research is focused around building Smart Robotic Assistants. To be smart, robots need to have the following capabilities- plan intelligently, learn from previous experiences, operate safely, ask for help, and efficiently interact with humans. These capabilities require robots to make complex decisions using AI. The students at RROS are working on Physics aware Artificial Intelligence for Smart Robotic Assistants.

Motion Planning and Self-Directed Learning for High Degrees Of Freedom Robotic Systems

Ariyan's research is focused on motion planning and self-directed learning algorithms that enable robots to program themselves. He has developed setup planning, point-to-point trajectory planning, and constrained trajectory planning algorithms using sampling, search, and optimization based methods. He has also developed real-time self-directed learning algorithms for parameter selection. The developed motion planning and self-directed learning algorithms have been verified through practical applications, including robotic cleaning and finishing.

Planning for Mobile Manipulators

Shantanu's current research is focused on trajectory planning for pick-up and transport operations of objects in a time-optimal manner. Essentially, the research involves generating motions resulting in object grasping while moving using a mobile manipulator. He has developed active learning based models for determining the probability of successful grasping. These models are used in non-linear programming for determining time-optimal trajectories. Shantanu has also worked on generating global trajectories for mobile manipulators using search algorithms object transportation operations. Further, he has also worked on planning for bi-manual mobile manipulation.

Policy Synthesis for Autonomous Multi-Robot Systems and Problems Related To Human Commanding Robots for Military Applications

Sarah Al-Hussaini

Sarah's research is focused on physics-aware decision-making for autonomous multi-agent systems operating in large-scale military and disaster-relief missions. It includes focuses on resiliency against uncertainties and stochastic failures, state estimation and recovery behaviors. She has worked on policy synthesis framework in the context of robot rescue in military applications. Sarah is also studying some problems related to human-robot teaming in such missions. She is currently looking into consistency checking of human-provided instructions to multi-robot teams in urgent, dangerous and time-sensitive scenarios.

Contingency planning for multi-robot missions

Shaurya Shriyam

Shaurya's current research is focused on leveraging formal methods and model checking tools for multi-robot mission planning in the presence of contingencies. The major hiccup to the deployment of autonomous systems is the guarantee of safety which maybe provided by simplifying the multi-robot system to a formal transition model and assessing the adequacy of its contingency resolution strategies. He has also developed scheduling algorithms to minimize mission makespan while accounting for the adverse impact of contingencies on multi-robot task allocation problems.

Trajectory Planning for Robots in Complex Environments

Pradeep Rajendran

Pradeep is interested in planning problems with complex environments, high-DOF robots and significant influence of uncertainties. His current research is focused on trajectory planning for manipulators and mobile robots. Specifically, he has developed trajectory planning algorithms for robots operating in congested and dynamic environments subject to sensing and motion uncertainty. These algorithms have been validated through physical experiments on Unmanned Surface Vehicles (USVs). He has also developed reactive trajectory planning for Unmanned Ground Vehicles (UGVs). In the area of manipulators, he has developed point-to-point trajectory planning algorithms for high-DOF robots in cluttered environments.

A multi-robot cell to perform sheet lamination-based additive manufacturing

Process Planning for Advanced Manufacturing using Robotic Systems

Prahar M. Bhatt

Artificial Intelligence (AI) is expected to revolutionise manufacturing. The current generation of AI-based advanced manufacturing technology has overcome many limitations of traditional manufacturing. However, the current advanced manufacturing technology still needs many improvements. Prahar's research focuses on how smart robotics can be used to realise advanced manufacturing. It deals with the motion planning, machine learning, etc. for executing manufacturing processes such as laminated object manufacturing, fused deposition modelling, and surface finishing with the aid of robotic manipulators.

Automation of Carbon Fiber Layup

Rishi K. Malhan

Composites are widely used to obtain high strength to weight ratios. Manufacturing of large simple geometries has been automated. Tape and fiber layup machines with large dispensing heads are used as a solution. The current automation solutions do not extend to small-medium and complex geometries. Hand layup of manually draping several carbon fiber prepreg sheets remains the only feasible process.

Rishi is working towards automating the hand layup process. The aim is to develop the computational foundations for manipulating and draping a flexible material like prepreg. Objective is to have the system capable of using the mold geometry and automatically generate robot placement and trajectory instructions. He is integrating inspection system for online robot trajectory correction. During this process, part quality is maintained as per manufacturing standards of the industry. The system operates under human supervision and assures quality.

Path Planning for Time-varying Marine Environments

The available free space in marine environment changes over time as a result of tides, environmental restrictions, and weather. As a result of these considerations, the free space region in marine environments needs to be dynamically generated and updated. This changes in environment have to considered during the computation of the long-distance path for Unmanned Surface Vehicles (USVs).

Brual is developing a long distance path planner for USVs that compute optimal paths using A* search on visibility graphs defined over quadtrees.

Resilient, Physics-Aware Decision Making for Autonomous Systems

Jason Gregory

In disaster-relief and military applications, such as exploration, search and rescue, and intelligence, surveillance, and reconnaissance missions, there is a ubiquitous threat of system and sub-system failure. Left unaccounted for, failure can have a catastrophic effect on mission performance and, in some cases, prevent operations altogether.

Jason's research focuses on improving resiliency and physics-aware decision making capabilities for autonomous, single- and multi-agent systems operating in these types of complex, real-world environments. He is specifically interested in characterizing, planning for, and mitigating the impact of system failure as it relates to fielded, human-agent teams. Jason has explored introspective approaches for autonomous capability quantification and enhanced detection and prediction of system failure. He has also developed policy generation techniques for online decision making to enable smart robotic rescue. Finally, Jason is investigating heuristics for multi-agent task allocation problems that, like in the real world, include actions with non-Markovian properties and probabilistic failure.