Aaron M. Dollar’s research while affiliated with Yale University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (279)


Fig. 1. A digital image of an herbarium specimen: the plant and original label containing the original metadata information, as well as additional reference information for the digital image: a ruler for scale and a color reference grid
Fig. 2. Overview of the confidence-based workflow: by only considering labels over a certain probability threshold, we increase the final accuracy of the model at the cost of coverage on the overall dataset (red: wrong label, green: true label, gray: rejected label)
Fig. 3. Rejection/Accuracy curves for all four models. Associated labels: (a) Budding, (b) Flowering, (c) Fruiting, (d) Reproductive.
Fig. 4. DoY estimation error and number of empty species estimates as a function of the confidence threshold
Fig. 5. Comparison of study ground-truth and AI estimate, per species, with a confidence threshold of 0.5. Blue: study ground truth with mean/std, red: AI estimate with mean/std.

+6

Improving the accuracy of automated labeling of specimen images datasets via a confidence-based process
  • Preprint
  • File available

November 2024

·

17 Reads

Quentin Bateux

·

Jonathan Koss

·

Patrick W. Sweeney

·

[...]

·

Aaron M. Dollar

The digitization of natural history collections over the past three decades has unlocked a treasure trove of specimen imagery and metadata. There is great interest in making this data more useful by further labeling it with additional trait data, and modern deep learning machine learning techniques utilizing convolutional neural nets (CNNs) and similar networks show particular promise to reduce the amount of required manual labeling by human experts, making the process much faster and less expensive. However, in most cases, the accuracy of these approaches is too low for reliable utilization of the automatic labeling, typically in the range of 80-85% accuracy. In this paper, we present and validate an approach that can greatly improve this accuracy, essentially by examining the confidence that the network has in the generated label as well as utilizing a user-defined threshold to reject labels that fall below a chosen level. We demonstrate that a naive model that produced 86% initial accuracy can achieve improved performance - over 95% accuracy (rejecting about 40% of the labels) or over 99% accuracy (rejecting about 65%) by selecting higher confidence thresholds. This gives flexibility to adapt existing models to the statistical requirements of various types of research and has the potential to move these automatic labeling approaches from being unusably inaccurate to being an invaluable new tool. After validating the approach in a number of ways, we annotate the reproductive state of a large dataset of over 600,000 herbarium specimens. The analysis of the results points at under-investigated correlations as well as general alignment with known trends. By sharing this new dataset alongside this work, we want to allow ecologists to gather insights for their own research questions, at their chosen point of accuracy/coverage trade-off.

Download

Chain-based lattice printing for efficient robotically-assembled structures

November 2024

·

24 Reads

·

1 Citation

Communications Engineering

Due to the nature of their implementation, nearly all low-level fabrication processes produce solidly filled structures. However, lattice structures are significantly stronger for the same amount of material, resulting in structures that are much lighter and more materially efficient. Here we propose an approach for fabricating lattice structures that echoes 3D printing techniques. In it, a modular chain of specially designed links is “extruded” onto a substrate to produce various lattices configurations depending on the chosen assembly algorithm, ranging from rigid regular lattices with nodal connectivity of 12, octet-truss, to significantly less dense configurations. Compared to conventional additive manufacturing methods, our approach allows for efficient use of nearly any material or combination of materials to construct lattices with programmed arrangements. We experimentally demonstrate that a 3x3x2 lattice structure (287 total links) is fabricated in 27 minutes via a modified robotic arm and can support approximately 1000 N in compression testing.







Transradial Amputee Reaching: Compensatory Motion Quantification Versus Unaffected Individuals Including Bracing

May 2024

·

16 Reads

·

2 Citations

IEEE Transactions on Medical Robotics and Bionics

Joint absence in people with upper-limb-difference leads to compensatory motions. Such compensation has long been a topic of study, but typically only for a single object/user layout, which is unlikely to spatially generalize. We seek to understand how motion varies over a planar workspace for different target orientations and wrist mobility conditions. We therefore present a study that records arm and torso pose during grasping of 49 equally spaced cylindrical targets. Furthermore, we seek to validate the research practice of using wrist-immobilizing bypass sockets on able-bodied participants to simulate prostheses without wrists. Participants were 2 transradial amputees and 7 able-bodied individuals who conducted the study with and without wrist braces, generating 2450 trajectories. Heat-maps illustrate variation over the workspace in Mean Joint Angle, Range of Joint Motion and Distance Travelled by Body Segment. Results indicate that greater wrist restriction primarily exacerbated shoulder internal rotation and elbow flexion, not the trunk. We observed that bypass sockets do not fully simulate amputee behavior. Furthermore, amputee reaching with their intact limb is different to the reaching motion of normative participants, implying that transradial limb-difference affects both sides of the body. Differences in participant behavior were also observed between horizontal and vertical target orientations.


FIGURE 3 (A) Seven energy maps are shown for seven actuation input combinations for a simple planar hand with two revolute fingers manipulating a square object in the plane. The commanded position (actuation input) to each finger is written in radians next to its joint. This hand is used as an example since it is easier to visualize a planar system's energy map than that of a spatial system. The gradient of each energy map is calculated locally at the object's drawn location, and is represented by a red vector. These vectors show how an object will move given its current location with the written actuation input to the system. (B) We find the largest ball inscribed inside the convex hull formed by the tips of the gradient vectors, which is a measure of manipulability of a hand-object system in a specific configuration. The example shown here is only three-dimensional for the purposes of illustration, but in this work we are actually finding 6-dimensional hyperspheres inscribed in the object's spatial configuration space. (C) The manipulability of the Allegro Hand with a cube is calculated over a grid and shaded accordingly. The model predicts that the hand is only capable of achieving high levels of dexterity (green) where the thumb and index fingers meet.
FIGURE 4 Eleven hands were simulated to manipulate cubes in this work throughout their workspace. The kinematic topologies for are shown for the following simulated hands: (A) underactuated and fully actuated T42; (B) Allegro Hand; (C) underactuated and fully actuated Model O; (D) underactuated and fully actuated Model Q; (E) H1; (F) H2; (G) Model (B,H) simulation world frame; (I) simulated cubes; (J) workspace grid (it is centered with the palm and raised one-half cube side length along the z-axis).
FIGURE 5 Top Panel: the hands simulated in this work are shown and sorted by the number of actuators. An 'F' in a hand's name indicates that it is the fully actuated version of an underactuated hand from Yale OpenHand. Bottom Panel: The same hands are now sorted by their performance based on the manipulation metric M avg described in section II.F.
FIGURE 6 A rendering of the Model B hand. It is comprised of four fingers-two pairs of identical opposing fingers. One set is prismatic, the other rotates about the axis of the palm. Each finger has a proximal and distal link. The distal links of each finger interdigitate with the opposing finger.
Robust whole-hand spatial manipulation via energy maps with caging, rolling, and sliding

November 2023

·

42 Reads

Frontiers in Robotics and AI

Humans regularly use all inner surfaces of the hand during manipulation, whereas traditional formulations for robots tend to use only the tips of their fingers, limiting overall dexterity. In this paper, we explore the use of the whole hand during spatial robotic dexterous within-hand manipulation. We present a novel four-fingered robotic hand called the Model B, which is designed and controlled using a straight-forward potential energy-based motion model that is based on the hand configuration and applied actuator torques. In this way the hand-object system is driven to a new desired configuration, often through sliding and rolling between the object and hand, and with the fingers “caging” the object to prevent ejection. This paper presents the first ever application of the energy model in three dimensions, which was used to compare the theoretical manipulability of popular robotic hands, which then inspired the design of the Model B. We experimentally validate the hand’s performance with extensive benchtop experimentation with test objects and real world objects, as well as on a robotic arm, and demonstrate complex spatial caging manipulation on a variety of objects in all six object dimensions (three translation and three rotation) using all inner surfaces of the fingers and the palm.


The CLaP System: Chain-based Lattice Printing for Efficient Robotically-Assembled Structures

November 2023

·

330 Reads

Due to the nature of their implementation, nearly all low-level fabrication processes produce solidly filled structures. However, lattice structures are significantly stronger for the same amount of material, resulting in structures that are much lighter and more materially efficient. In this paper we propose an approach for fabricating lattice structures that echoes commercially successful additive manufacturing/3D printing techniques. In it, a modular chain of specially designed links is “extruded” onto a substrate to produce various lattice configurations depending on the chosen assembly algorithm, ranging from one of the strongest known rigid lattices, the octet-truss, to significantly less dense configurations. Like 3D printing, the process allows material to be compactly stored before being deployed into structures of nearly arbitrary geometry, but unlike it, the approach allows for the use of nearly any material or combination of materials. We demonstrate the concept experimentally with a nearly 300-link chain that is fed from a spool and autonomously laid down onto programmed lattice arrangements via a robotic arm modified for the task. In the current prototype implementation, ~300-link structures are fabricated in 27 minutes, and a 3x3x2 lattice structure (287 total links) is shown to support approximately 1000N in compression testing.


Citations (67)


... In a publication in Communications Engineering, Zhe Xu and Aaron Dollar introduce "Chain-Based Lattice Printing (CLaP)", an approach leveraging a modular chain of interlinked nodes to create efficient lattice structures 16 . The strategy uses a robot extruder, similar to a 3D printer, but one that extrudes a pre-existing chain of struts. ...

Reference:

Editors’ Choice 2024
Chain-based lattice printing for efficient robotically-assembled structures

Communications Engineering

... Most of the existing work on multi-agent task persistence with static charging stations assumes either a dedicated charging station for each agent [10]- [12], [21] or a single charging station that can cater to multiple agents simultaneously [13], [14]. Additionally, some studies address the scenarios where the number of charging stations available is fewer than the number of robots [15]- [17]. They do this by either altering the nominal paths to visit charging stations along the way [15] or strategically placing the shared charging resources [16], or by imposing a constraint in the optimization problem to ensure that only * The authors would like to acknowledge the support of the National Science Foundation (NSF) under grant no. ...

Energy-Aware Ergodic Search: Continuous Exploration for Multi-Agent Systems with Battery Constraints
  • Citing Conference Paper
  • May 2024

... The dataset used in this study is from [14], in which the workspace in front of the subjects was discretely sampled by a 7×7 square grid at 300mm intervals, horizontally and vertically. The subjects -seven non-disabled, righthanded adults -were asked to reach to and grasp each of the 49 cylindrical targets (one at a time) either orientated horizontally or vertically, depending on the placement of the grid. ...

Transradial Amputee Reaching: Compensatory Motion Quantification Versus Unaffected Individuals Including Bracing
  • Citing Article
  • May 2024

IEEE Transactions on Medical Robotics and Bionics

... While an object's position change can be visualized using a position trace, it is more challenging to visualize its orien-tation over time. Several works seek to visualize orientations on position traces using color [9] or coordinate frames [10], [11], which can be unintuitive or visually cluttered. An alternative body of literature represents orientations in quaternions and seeks ways to visualize quaternion traces, e.g., by projecting quaternions to a 3D space [12]. ...

An Analysis of Unified Manipulation with Robot Arms and Dexterous Hands via Optimization-based Motion Synthesis
  • Citing Conference Paper
  • May 2023

... The integration of open-source hardware in educational robotics has been proposed as a potential solution (Heradio et al., 2018;Patel et al., 2023) as it holds a promise of accessible, customisable and transparent development. However, being open-source does not guarantee universal accessibility, as the fabrication might be costly or call for techniques normally beyond the reach of educators or self-guided learners. ...

Open Robot Hardware: Progress, Benefits, Challenges, and Best Practices

IEEE Robotics & Automation Magazine

... A learning-based dynamics model typically involves an encoder that maps the visual observation to a scene representation, and a predictive model predicts the representation's evolution given an external action. Different choices of scene representations (e.g., latent vectors [21,20,34], object-centric [71,15] or keypoint representations [41,38,67]) imply different expressiveness and generalization capabilities. Therefore, it is of critical importance to think carefully about what scene representation to use for a given task. ...

Dynamical Scene Representation and Control with Keypoint-Conditioned Neural Radiance Field
  • Citing Conference Paper
  • August 2022

... It records and translates the surface electromyography (sEMG) on the amputation stump into commands to realize the direct neural control on the mechanical body of the prosthesis. In recent years, with the development of engineering technologies, the dexterity of humanoid prosthetic hands and the performance of integrated sensors have been continuously improved [3], [4], [5], [6]. However, the neural control interface and human-machine interaction performance of prostheses are inadequate to match dexterous operation [7], [8]. ...

The Yale MyoAdapt Hand: A Highly Functional and Adaptive Single Actuator Prosthesis
  • Citing Article
  • August 2022

IEEE Transactions on Medical Robotics and Bionics

... However, these methodologies have typically failed to consider the curvier and more eccentric displacements observed in natural movements, as documented in [38]. Another approach involving the determination of predefined joint trajectories derived from natural movements performed in daily activities, which users can choose for task execution, has demonstrated its advantage over traditional myoelectric control methods [39]. Nevertheless, the available trajectories are constrained to those predefined within the system, far from encompassing the diversity of movements achievable with our upper arm. ...

Trajectory Control – An Effective Strategy for Controlling Multi-DOF Upper Limb Prosthetic Devices

IEEE transactions on neural systems and rehabilitation engineering: a publication of the IEEE Engineering in Medicine and Biology Society

... In-Hand Manipulation. In-hand manipulation has been studied for decades [16], [17], [18], [19], [20], [21], [22], [23], [24], [25] in classic robotics. More recently, learningbased methods have achieved significant progress [1], [3], [4], [26], [27], [28], [29], [30]. ...

Complex In-Hand Manipulation Via Compliance-Enabled Finger Gaiting and Multi-Modal Planning
  • Citing Article
  • April 2022

IEEE Robotics and Automation Letters