Modeling the topological structure underlying decision making of rational agents in structured navigation environments may result in significantly safer execution and improved performance in challenging environments such as unsignalized intersections. We are developing a decentralized, braids-based motion planner for safe navigation in such environments. [arXiv][journal article under review]
MuSHR is an open-source, low-cost mobile robot platform developed at the Personal Robotics Lab and the main platform in my multi-robot coordination line of work. I am leading the MuSHR team that maintains the platform, extends its capabilities and conducts research with it. MuSHR is also the experimental platform that I am using in my undergraduate robotics course.
We make use of topological braids as symbols describing distinct classes of multi-agent navigation behavior. This representation allows a robot to infer the unfolding multi-agent dynamics and act legibly towards simplifying inference for others. Our framework results in accelerated implicit coordination among agents which leads to high time efficiency. [WAFR '16][IROS '17][IJRR '19]
To overcome the high computational load of reasoning about multi-agent navigation strategies, we developed a motion planner (the Social Momentum) that only looks at the pairwise collision avoidance intentions between the robot and other agents. Our planner is inspired by the physical quantity of Angular Momentum, which we use as a heuristic to label pairwise avoidance intentions. An online, video-based user study revealed evidence that our framework results in legible behaviors in multi-agent environments with multiple non-communicating navigating agents. [HRI '18][journal article under review]
We found that the topological foundation of Social Momentum is the topological invariant of the winding number. We used this understanding to construct dynamic models that generate multi-agent trajectories from topological specification. We built a planner that makes use of this mechanism to generate topologically robust predictions online. Our planner was shown to allow for rapid adaptation to heterogeneous agents and agents with changing intentions. [WAFR '18][invited journal article under review]
Towards understanding human versatility and adaptability in manipulation tasks, we developed a novel robotic system for chopsticks-based telemanipulation. Through a lab study with 25 participants, we found that our system enabled users to successfully grasp a wide variety of objects and observed that users quickly adapted to our teleoperation interface. Our system may prove useful in providing high-quality demonstrations for challenging manipulation tasks. [IROS '20]
We envision a collaborative, multi-robot system, capable of reconfiguring a set of objects in a cluttered environment without relying on explicit communication. As a first step, we are focusing on the task of block pushing, considering a single nonholonomic pusher and a single block. As a testbed, we consider the MuSHR platform, augmented with a flat bumper.
My Diploma thesis focused on the design of grasp planning optimization schemes for multi-fingered robot hands. Taking as input the location and the physical properties of an object, as well as a series of different grasp quality criteria, my schemes generated optimal grasps (force distribution and hand posture). [ICRA '13][ICRA '14]
I have been involved in a series of projects on the design of robot hands and arms. I have looked at the mechanical and computational aspects of design towards producing affordable, highly-functional, anthropomorphic robotic and prosthetic hands and robot manipulators [IROS '14][IROS '15][IROS '15]
We conducted an in-lab experimental study in which 105 participants navigated next to a robot following 3 different navigation strategies. We collected participants' trajectories and their impressions of the robot's behavior through questionnaires. Highlights of our analysis include the low-acceleration paths that humans follow next to a robot running Social Momentum and the lack of evidence to support our expectation that humans would find teleoperation as more intelligent and humanlike. [HRI '19]
Imbuing robotic agents with an understanding of the mechanisms underlying human decision making in manipulation tasks may enable improved performance in joint manipulation tasks. As a testbed, we consider the application of packing, in which a human packs a container in collaboration with a robot manipulator. As a first step, we explore and analyze human preferences and strategies in packing tasks towards transferring our insights into the design of a decision making mechanism for a robot partner. [AI-HRI '19]
We explored the fundamental mathematical mechanisms underlying implicit communication in a series of different applications, including conversational implicature, legible collaborative manipulation and social navigation. We showed that by performing an unexpected action, an actor may achieve effective communication of its goal to non-explicitly-communicating observers. [HRI '17]