Source Themes

We propose B4P, a forest-based planning framework that simultaneously finds grasp configurations and feasible robot motions for object placement. Our bidirectional sampling-based approach builds start and goal forests to connect valid grasp-placement pairs, enabling superlinear speedup through inherent parallelism. This makes the framework scalable for redundant robot arms working in highly cluttered environments, addressing the sub-optimality introduced by traditional decoupled planning approaches.

We present Roller Rings (RR), a modular robotic attachment with active surfaces that enables dexterous in-hand manipulation when worn by robot and human hands. By angling the RRs such that their motions are not co-linear, we derive a general differential motion model showing that complete manipulation skill sets can be achieved with as few as 2 RRs through non-holonomic object motions. More RRs enhance dexterity with fewer constraints. Experiments on robot and human hands validate the RRs’ capability to manipulate arbitrary object shapes and provide dexterous in-hand manipulation.

We propose a novel object-centric planning paradigm for nonprehensile robot rearrangement manipulation. Our approach focuses on planning desired object motions, which are then realized via robot actions generated online. Experiments show that this paradigm generates more intuitive and efficient robot actions compared to traditional robot-centric approaches.

Controlling robotic in-hand manipulation is complex due to the varying physics and required system knowledge. One model, the inverse Jacobian, translates desired object motions to hand controls, but acquiring it without sophisticated system models is challenging. Our method uses a particle filter-based scheme to self-identify inverse Jacobians, enabling underactuated hands to stably grasp during self-identification movements. This approach requires no prior knowledge and learns the system’s inverse Jacobian through exploratory motions. Our system closely approximates the inverse Jacobian, performing manipulation tasks successfully. Experiments on a Yale Model O hand show sub-millimeter precision and real-time control up to 900Hz.

We propose a self-calibration framework that allows robots to autonomously calibrate themselves into their workspaces using built-in force-torque sensors. Through compliant exploratory actions, the robot interactively explores the environment’s geometries to estimate spatial relationships without relying on external sensors or human intervention. Experiments validate the effectiveness of our approach in accurately establishing robot-environment calibration.

We introduce RISeg, a robot interactive perception framework that significantly improves unseen object segmentation in cluttered scenes. RISeg identifies regions of segmentation uncertainty from a static image-based model, introduces object motion through minimal robot interactions, and matches spatial twists of randomly sampled object frames to group them. This allows accumulated segmentation corrections without relying on object singulation. RISeg increases segmentation accuracy by 28.2% over static methods on real cluttered tabletop scenes.

We propose UNO Push, a unified framework for precise nonprehensile object pushing that jointly addresses system modeling, action generation, and control. Our approach approximates a system transition function via non-parametric learning using only a small number of exploratory actions, and integrates it with model predictive control. The approximated functions can be robustly transferred across novel objects and online updated for continuous improvement. Experiments on a real robot platform demonstrate that UNO Push is a light-weight and highly effective approach, achieving millimeter-level precision on novel objects without relying on a priori system information or large datasets.

This work presents a method for achieving complete SO(3) finger gating control of grasped objects against gravity, using a manipulation planner that operates via orthogonal safe modes of a compliant, underactuated hand absent of tactile sensors or joint encoders. The method takes advantage of system compliance to allow the hand to more easily switch contacts while maintaining a stable grasp, and uses a low-latency 6D pose object tracker for online feedback and adaptive recovery from trajectory deviations. The method is demonstrated on a real robot manipulating both convex and non-convex objects, and represents a valuable step towards realizing true robot in-hand manipulation capabilities.

We propose a solution to the problem of herding by caging: given a set of mobile robots (called herders) and a group of moving agents (called sheep), we guide the sheep to a target location without letting them escape from the herders along the way. We model the interaction between the herders and the sheep by defining virtual “repulsive forces” pushing the sheep away from the herders. This enables the herders to partially control the motion of the sheep. We formalize this behavior topologically by applying the notion of caging, a concept used in robotic manipulation. We demonstrate that our approach is provably correct in the sense that the sheep cannot escape from the robots under our assumed motion model. We propose an RRT-based path planning algorithm for herding by caging, demonstrate its probabilistic completeness, and evaluate it in simulations as well as on a group of real mobile robots.

The process of modeling a series of hand-object parameters is crucial for precise and controllable robotic in-hand manipulation because it enables the mapping from the hand’s actuation input to the object’s motion to be ob-tained. Without assuming that most of these model parameters are known a priori or can be easily estimated by sensors, we focus on equipping robots with the ability to actively self-identify necessary model parameters using minimal sensing. Here, we derive algorithms, on the basis of the concept of virtual linkage-based representations (VLRs), to self-identify the underlying mechanics of hand-object systems via exploratory manipulation actions and probabilistic reasoning and, in turn, show that the self-identified VLR can enable the control of precise in-hand ma-nipulation. To validate our framework, we instantiated the proposed system on a Yale Model O hand without joint encoders or tactile sensors. The passive adaptability of the underactuated hand greatly facilitates the self-identification process, because they naturally secure stable hand-object interactions during random exploration. Relying solely on an in-hand camera, our system can effectively self-identify the VLRs, even when some fingers are replaced with novel designs. In addition, we show in-hand manipulation applications of handwriting, marble maze playing, and cup stacking to demonstrate the effectiveness of the VLR in precise in-hand manipulation control.