David Cordova Bulens

Dr. David Cordova Bulens

orcid

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 November 2022 in Expert Systems
Verified
Journal Article
Open access
QWQiang Wang
Qiang Wang
FSFrancisco Roldan Sanchez
Francisco Roldan Sanchez
RMRobert McCarthy
Robert McCarthy
DBDavid Cordova Bulens
David Cordova Bulens
KMKevin McGuinness
Kevin McGuinness
SRStephen Redmond
Stephen Redmond
This paper describes a deep reinforcement learning (DRL) approach that won Phase 1 of the Real Robot Challenge (RRC) 2021, and then extends this method to a more difficult manipulation task. The RRC consisted of using a TriFinger robot to manipulate a cube along a specified positional trajectory, but with no requirement for the cube to have any specific orientation. We used a relatively simple reward function, a combination of a goal‐based sparse reward and a distance reward, in conjunction with Hindsight Experience Replay (HER) to guide the learning of the DRL agent (Deep Deterministic Policy Gradient [DDPG]). Our approach allowed our agents to acquire dexterous robotic manipulation strategies in simulation. These strategies were then deployed on the real robot and outperformed all other competition submissions, including those using more traditional robotic control techniques, in the final evaluation stage of the RRC. Here we extend this method, by modifying the task of Phase 1 of the RRC to require the robot to maintain the cube in a particular orientation, while the cube is moved along the required positional trajectory. The requirement to also orient the cube makes the agent less able to learn the task through blind exploration due to increased problem complexity. To circumvent this issue, we make novel use of a Knowledge Transfer (KT) technique that allows the strategies learned by the agent in the original task (which was agnostic to cube orientation) to be transferred to this task (where orientation matters). KT allowed the agent to learn and perform the extended task in the simulator, which improved the average positional deviation from 0.134 to 0.02 m, and average orientation deviation from 142° to 76° during evaluation. This KT concept shows good generalization properties and could be applied to any actor‐critic learning algorithm.

1 March 2018 in Journal of Neurophysiology
Verified
Journal Article
Open access
PDF file
DBDavid Cordova Bulens
David Cordova Bulens
FCFrédéric Crevecœur
Frédéric Crevecœur
JTJean-Louis Thonnard
Jean-Louis Thonnard
PLPhilippe Lefèvre
Philippe Lefèvre
Bimanual tasks involve the coordination of both arms, which often offers redundancy in the ways a task can be completed. The distribution of control across limbs is often considered from the perspective of handedness. In this context, although there are differences across dominant and nondominant arms during reaching control ( Sainburg 2002 ), previous studies have shown that the brain tends to favor the dominant arm when performing bimanual tasks ( Salimpour and Shadmehr 2014 ). However, biomechanical factors known to influence planning and control in unimanual tasks may also generate limb asymmetries in force generation, but their influence on bimanual control has remained unexplored. We investigated this issue in a series of experiments in which participants were instructed to generate a 20-N force with both arms, with or without perturbation of the target force during the trial. We modeled the task in the framework of optimal feedback control of a two-link model with six human-like muscles groups. The biomechanical model predicted a differential contribution of each arm dependent on the orientation of the target force and joint configuration that was quantitatively matched by the participants’ behavior, regardless of handedness. Responses to visual perturbations were strongly influenced by the perturbation direction, such that online corrections also reflected an optimal use of limb biomechanics. These results show that the nervous system takes biomechanical constraints into account when optimizing the distribution of forces generated across limbs during both movement planning and feedback control of a bimanual task. NEW & NOTEWORTHY Here, we studied a bimanual force production task to examine the effects of biomechanical constraints on the distribution of control across limbs. Our findings show that the central nervous system optimizes the distribution of force across the two arms according to the joint configuration of the upper limbs. We further show that the underlying mechanisms influence both movement planning and online corrective responses to sudden changes in the target force.

13 December 2023 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
QWQiang Wang
Qiang Wang
RMRobert McCarthy
Robert McCarthy
DBDavid Cordova Bulens
David Cordova Bulens
FSFrancisco Roldan Sanchez
Francisco Roldan Sanchez
KMKevin McGuinness
Kevin McGuinness
SRStephen Redmond
Stephen Redmond
This letter presents our solution for the Real Robot Challenge III 1 , aiming to address dexterous robotic manipulation tasks through learning from offline data. In this competition, participants were given two types of datasets for each task: expert and mixed. Each expert dataset is collected by a high-skill policy, whereas the mixed dataset is collected using both expert and non-expert policies. We found that the vanilla behavioural cloning (BC) can learn a very proficient policy with minimal human intervention when trained on expert datasets. Notably, BC outperformed even the most advanced offline reinforcement learning (RL) algorithms. However, when applied to mixed datasets, the performance of BC deteriorates; similarly, the performance of offline RL algorithms is also less than satisfactory. Upon examining the provided datasets, it was apparent that each mixed dataset contained a significant proportion of expert data, which should enable the training of a proficient BC agent. However, the expert data is not labelled in the datasets. As a result, we propose a classifier to identify the pattern of the expert behaviour within a mixed dataset and then utilize it to isolate the expert data. To further boost the BC performance, we take advantage of the geometric symmetry of the arena to augment the training dataset through mathematical transformations. Ultimately, our submission outperformed that of all other participants.

26 December 2023 in IEEE Robotics and Automation Letters
Verified
Journal Article
Closed access
QWQiang Wang
Qiang Wang
PUPablo Martinez Ulloa
Pablo Martinez Ulloa
RBRobert Burke
Robert Burke
DBDavid Cordova Bulens
David Cordova Bulens
SRStephen Redmond
Stephen Redmond
The ability to detect slip, particularly incipient slip, enables robotic systems to take corrective measures to prevent a grasped object from being dropped. Therefore, slip detection can enhance the overall security of robotic gripping. However, accurately detecting incipient slip remains a significant challenge. In this letter, we propose a novel learning-based approach to detect incipient slip using the PapillArray (Contactile, Australia) tactile sensor. The resulting model is highly effective in identifying patterns associated with incipient slip, achieving a detection success rate of 95.6% when tested with an offline dataset. Furthermore, we introduce several data augmentation methods to enhance the robustness of our model. When transferring the trained model to a robotic gripping environment distinct from where the training data was collected, our model maintained robust performance, with a success rate of 96.8%, providing timely feedback for stabilizing several practical gripping tasks. Our project website, https://sites.google.com/view/incipient-slip-detection , contains the code, datasets, and video demos for this letter.

14 July 2023 in IEEE Sensors Journal
Verified
Journal Article
Open access
OLOlivia Grace Leslie
Olivia Grace Leslie
DBDavid Cordova Bulens
David Cordova Bulens
PUPablo Martinez Ulloa
Pablo Martinez Ulloa
SRStephen Redmond
Stephen Redmond
This paper proposes an optical-based tactile sensor design concept, which uses a light angle and intensity sensor to infer force and displacement from deformations of a silicone pillar. The proposed design uses a simple, low-cost fabrication method with an overall small-scale form factor. The sensor can measure 3D displacement, 3D force, and vibration. The overall displacement estimation error (mean ± SD) in the X, Y, and Z axes was -10.8 ± 52.1 μm, 15.4 ± 66.5 μm and 1.8 ± 17.6 μm, respectively, over a full-scale lateral displacement of 1 mm radius in X and Y and 2.2 mm compression in Z. The overall force estimation error (mean ± SD) was -8.5 ± 47.6 mN, -8.5 ± 49.0 mN and -28.0 ± 92.6 mN for a full-scale force of approximately 2 N in X or Y, and 6 N in Z. Sensitivity to vibrations in the range of 10-950 Hz was also evaluated showing good sensitivity over this entire range. This new sensing approach could be of benefit in robotic manipulation applications, as it could be easily arrayed and/or integrated into the fingers of a robotic gripper to sense slip events and measure load and grip forces and torques.

5 December 2023 in Sensors
Verified
Journal Article
Open access
PDF file
OLOlivia Grace Leslie
Olivia Grace Leslie
DBDavid Cordova Bulens
David Cordova Bulens
SRStephen Redmond
Stephen Redmond
Real-time multi-axis distributed tactile sensing is a critical capability if robots are to perform stable gripping and dexterous manipulation, as it provides crucial information about the sensor–object interface. In this paper, we present an optical-based six-axis tactile sensor designed in a fingertip shape for robotic dexterous manipulation. The distributed sensor can precisely estimate the local XYZ force and displacement at ten distinct locations and provide the global XYZ force and torque measurements. Its compact size, comparable to that of a human thumb, and minimal thickness allow seamless integration onto existing robotic fingers, eliminating the need for complex modifications to the gripper. The proposed sensor design uses a simple, low-cost fabrication method. Moreover, the optical transduction approach uses light angle and intensity sensing to infer force and displacement from deformations of the individual sensing units that form the overall sensor, providing distributed six-axis sensing. The local force precision at each sensing unit in the X, Y, and Z axes is 20.89 mN, 19.19 mN, and 43.22 mN, respectively, over a local force range of approximately ±1.5 N in X and Y and 0 to −2 N in Z. The local displacement precision in the X, Y, and Z axes is 56.70 μm, 50.18 μm, and 13.83 μm, respectively, over a local displacement range of ±2 mm in the XY directions and 0 to −1.5 mm in Z (i.e., compression). Additionally, the sensor can measure global torques, Tx, Ty, and Tz, with a precision of of 1.90 N-mm, 1.54 N-mm, and 1.26 N-mm, respectively. The fabricated design is showcased by integrating it with an OnRobot RG2 gripper and illustrating real-time measurements during in simple demonstration task, which generated changing global forces and torques.