Contact: akshaybh(at)umich.edu

Hi! My name is Akshay. I am an engineer and a researcher in robotics. My research focuses on understanding how humans physically interact with robots and on designing robots that can safely and efficiently collaborate with humans.

I am currently pursuing a PhD in Mechanical Engineering at the University of Michigan where I am working in the HaptiX Lab under the supervision of Prof. Brent Gillespie. My PhD work focuses on the development of steering assistance and augmentation systems for semi-autonomous vehicles.

Recent News...

About

I am a PhD candidate in Mechanical Engineering at the University of Michigan Ann Arbor. My research focuses on studying the human behavior when humans physically interact with machines that are partially (or fully) autonomous, and using the insights driven from such studies to design and evaluate human-machine interfaces that would enable efficient and safe human-machine collaboration. My research blends methods from analytical dynamics, human factors, motor control, and haptics, and crosses the boundary between various disciplines such as engineering, biomechanics, ergonomics, and psychophysics.

My current work specifically focuses on understanding the driver behavior in partially automated vehicles to design steering assistance systems that can improve both the driver comfort and safety. My current research interests include physical human-machine interaction, modeling and control of driver assistance systems, and design of haptic interfaces.

I am originally from New Delhi, India. I received a B.E. degree in Manufacturing Processes and Automation Engineering from University of Delhi (India) in 2015, and an M.S.E. degree in Mechanical Engineering from University of Michigan Ann Arbor in 2018.

Research

Control Sharing in Emergency Situations

Automation can mitigate or avoid accidents in emergency scenarios where driver inputs might be insufficient. However, during an automation failure, a fully autonomous vehicle might render the human incapable of avoiding a collision. Haptic shared control (HSC) allows both the human driver and automation system to cover for each other's faults. The automation system in HSC is designed to collaborate with human with a fixed mechanical impedance. On the other hand, the human driver can modulate their impedance through muscle action to override or acquiesce to the automation system. However, we think that a trade-off exists between the automation impedance and fault tolerance provided by the driver.


Intuitive Interfaces for Vehicle Steering

Humans are increasingly asked to collaborate with robots in everyday tasks. But it should not be up to humans to learn how to work with robots. Robots must be designed to support safe and effective human-robot collaboration. A high-performing human-robot team is built on effective communication of each agent’s intent and control authority. Haptic interaction can facilitate such communication. For instance, in haptic shared control of steering, human driver and automation can read each other’s intent (steering angle) and control authority (impedance) by feeling torque on the steering wheel. This results in significant improvements in human-automation team performance. In this spirit, I am now building a grip force sensing and shape changing steering wheel.


Continuous and Discrete Control Sharing

Before fully autonomous vehicles become a reality, it is required that the control between the human and automatic drivers be first shared in a way that allows either driver to cover for the faults of the other. However, developing a human-automation control sharing scheme is a notoriously challenging problem because humans working alongside automation are prone to loss of skills, situation awareness, and mode awareness. To address this problem, I designed two types of control sharing schemes, called discrete and continuous schemes, and implemented them on a driving simulator.


Steering Feel on Uneven Roads

The torque experienced by a driver at the steering wheel, also called steering feel, couples the driver’s hands and arms to the vehicle and is critical to smooth and controlled vehicle maneuvers. A major portion of the steering torque feedback comes from the rack force which is defined as the net force transmitted from tires to the steering rack of a vehicle. As a result, estimates of rack force are used in a wide variety of advanced driver assist systems. Existing methods for rack force estimation are either only applicable for driving on flat roads or are susceptible to steering system disturbances. I used vehicle dynamics theory to build models that can produce disturbance-free rack force estimates for driving on arbitrarily uneven road surfaces.


Undergraduate Research

Teleoperation with Haptic Feeback

Teleoperated robotic surgery holds numerous benefits such as shorter patient recovery time, lower blood loss, and lower risk of tissue trauma and infection. Training surgeons directly on surgical robots is costly due to the limited availability of robots and the potential misuse, or overuse, of robotic parts that requires expensive replacements. Robotic surgery simulators have the potential to significantly reduce the time, cost, and resources required to train surgeons. During my undergrad, I developed a surgery simulator for an open-source surgical robot called Raven-II. I further used adaptive filter theory to develop an algorithm to reduce teleoperation tracking errors that arise due to transmission delays and data losses.

Intuitive Interfaces for Vehicle Steering

Humans are increasingly asked to collaborate with robots in everyday tasks. But it should not be up to humans to learn how to work with robots. Robots must be designed to support safe and effective human-robot collaboration. A high-performing human-robot team is built on effective communication of each agent’s intent and control authority. Haptic interaction can facilitate such communication. For instance, in haptic shared control of steering, human driver and automation can read each other’s intent (steering angle) and control authority (impedance) by feeling torque on the steering wheel. This results in significant improvements in human-automation team performance. In this spirit, I am now building a grip force sensing and shape changing steering wheel.

In my first design, the force sensing and shape change are realized by two fluid bladders attached to the steering wheel. Through the steering wheel both human and automation will be able to send a take over request: driver by squeezing the steering wheel and automation by governing fluid mass into the bladders. Moreover, each agent will be able to sense other agent’s take over request: driver through change in shape and automation through change in pressure. Automation can also communicate when the take over request is not granted (by keeping the wheel dilated) and can further adapt its impedance (in the orthogonal steering axis) in real-time to make the authority transitions smoother.

Control Sharing in Emergency Situations

Automation can mitigate or avoid accidents in an emergency situation where driver’s reaction speed is challenged. However, during an automation failure, a fully autonomous vehicle might render the human incapable of avoiding a collision. Haptic shared control (HSC) allows both the human driver and automation system to cover for each other's faults. The automation system in HSC is designed to collaborate with human with a fixed mechanical impedance. On the other hand, the human driver can modulate their impedance through muscle action to override or acquiesce to the automation system. However, we think that a trade-off exists between the automation impedance and fault tolerance provided by the driver. Setting the automation impedance high might allow automation to cover for human faults but might make it difficult for the driver to avoid the obstacles during automation faults. To test this, I designed and compared the performance of high and low impedance HSC at avoiding obstacles during emergency situations and automation faults.

During emergency situations, automation was designed to avoid the obstacles (intended automation), whereas during automation faults, automation was designed to false activate (adversarial automation). The results confirmed the hypothesis that high impedance HSC is significantly more effective at avoiding obstacles during emergency situations but is less effective at avoiding obstacles during automation faults. To combine the advantages of low and high impedance HSC I have started worked on the design of an adaptive impedance HSC that would assume a high impedance during emergency situations in which the automation has high confidence, and low impedance in situations in which the automation has low confidence.

Continuous and Discrete Control Sharing

Before fully autonomous vehicles become a reality, it is required that the control between the human and automatic drivers be first shared in a way that allows either driver to cover for the faults of the other. However, developing a human-automation control sharing scheme is a notoriously challenging problem because humans working alongside automation are prone to loss of skills, situation awareness, and mode awareness. To address this problem, I designed two types of control sharing schemes, called discrete and continuous schemes, and implemented them on a driving simulator.

In discrete control sharing, the human driver or automation system initiated a transition of control authority by a trigger (such as pressing a button) at discrete intervals of time. In the continuous control sharing scheme, called haptic shared control, the driver and the automation system continuously shared steering control through haptic feedback. I performed human subject experiments to compare the two schemes in a simulated driving scenario in which faults occured at fixed rates but at unpredictable times. The results of the experiments showed that haptic shared control resulted in higher driver awareness, reduced collisions and more efficient obstacle avoidance maneuvers in comparison to discrete control sharing conditions.

Steering Feel on Uneven Roads

The torque experienced by a driver at the steering wheel, also called steering feel, couples the driver’s hands and arms to the vehicle and is critical to smooth and controlled vehicle maneuvers. A major portion of the steering torque feedback comes from the rack force which is defined as the net force transmitted from tires to the steering rack of a vehicle. As a result, estimates of rack force are used in a wide variety of advanced driver assist systems. Existing methods for rack force estimation are either only applicable for driving on flat roads or are susceptible to steering system disturbances. I used vehicle dynamics theory to build models that can produce disturbance-free rack force estimates for driving on arbitrarily uneven road surfaces. I categorized road profile variations based on their frequencies and utilized tire models that could capture the tire force variations for such frequencies. I validated the estimation accuracy of my models by performing driving tests on an actual vehicle on test tracks with known or measured road profile variations.

Teleoperation with Haptic Feedback

Teleoperated robotic surgery holds numerous benefits such as shorter patient recovery time, lower blood loss, and lower risk of tissue trauma and infection. Training surgeons directly on surgical robots is costly due to the limited availability of robots and the potential misuse, or overuse, of robotic parts that requires expensive replacements. Robotic surgery simulators have the potential to significantly reduce the time, cost, and resources required to train surgeons. During my undergrad, I developed a surgery simulator for an open-source surgical robot called Raven-II. I used the Phantom Omni haptic device to teleoperate a model of the robot placed in a virtual environment, and to provide haptic feedback arising from the interaction between the virtual robot with the objects placed in the virtual environment. I further used adaptive filter theory to develop an algorithm to reduce the master-slave tracking errors that arise due to transmission delays and data losses during teleoperation. I implemented the algorithm on two Phantom devices to demonstrate its effectiveness.

Publications

Following is a selection of my publications. For a complete list of my papers, and a list of my presentations please refer to my CV.

Journal Article in Submission

  • Bhardwaj, A., Slavin, D., Walsh, J., Freudenberg, J., Gillespie, R.B. “Estimation and Decomposition of Rack Force for Driving on Uneven Roads". [Preprint]

Refereed Journal Article

  • Bhardwaj, A., Ghasemi, A., Zheng, Y., Febbo, H., Jayakumar, P., Ersal, T., Stein, J., Gillespie, R. B. "Who's the Boss? Arbitrating Control Authority between a Human Driver and Automation System". In Transportation Research F: Traffic Psychology and Behavior, Elsevier, 68 (2020), 144-160. [Paper]

Refereed Conference Papers

  • Bhardwaj, A., Lu, Y., Pan, S., Sarter, N., Gillespie, R.B. "The Effects of Driver Coupling and Automation Impedance on Emergency Steering Interventions". In IEEE International Conference on Systems, Man and Cybernetics 2020. (Accepted). [Preprint]

  • Bhardwaj, A., Slavin, D., Walsh, J., Freudenberg, J., Gillespie, R.B. “Rack Force Estimation for Driving on Uneven Road Surfaces". In the IFAC World Congress 2020. Young Author Prize Honorable Mention. [Preprint]

  • Izadi, V., Bhardwaj, A., Ghasemi, A. “Impedance Modulation for Negotiating Control Authority in a Haptic Shared Control Paradigm". In the American Control Conference 2020, IEEE. [Paper]

  • Bhardwaj, A., Gillespie, R.B., Freudenberg, J. “Estimating Rack Force due to Road Slopes for Electric Power Steering Systems". In the American Control Conference 2019, IEEE. [Paper]

  • Bhardwaj, A., Jain, A., Agarwal, V. “Preoperative planning simulator with haptic feedback for Raven-II surgical robotics platform". In Computing for Sustainable Global Development 2016, IEEE. [Paper]

  • Bhardwaj, A., Agarwal, V., Parthasarathy, H. “Bilateral teleoperation control using recursive least squares filter with forgetting factor". In India Conference (INDICON) 2015, IEEE. Best Paper in Control Winner. [Paper]

Teaching

I really enjoy teaching and mentoring students. Following is one of my recent experiences related to teaching. For details on my mentoring experience please refer to my CV.

ME 360: Modeling, Analysis and Control of Dynamic Systems (Teaching Assistant)

I was a teaching assistant (TA) for this junior/senior level undergraduate class (of 74 students) covering dynamic systems modeling and classical control techniques. As a TA, I was responsible for holding office hours and leading recitation classes every week. I assisted the course instructor in the preparation of exams, weekly quizzes and assignments and also prepared and delivered a lecture on "Bode Plotting".

In addition, I helped the instructor design new lab exercises based on a device, called the Cigar Box, that was previously only used in graduate level classes. The device consisted of a cigar box outfitted with two motorized linear potentiometers, three rotary potentiometers, and an Arduino. The cigar box was intended to give students hands-on experience in applying and observing the principles of control design using Matlab.


Service

I have served as a reviewer for several academic conferences and journals. At the department level, I have lead grad council committees, have served on student panels, and have volunteered in outreach activities. Following are the two activities I am currently involved in. For details on other activities please see my CV.

Greenhills School Research Mentoring (Research Host)

I am serving as a research host for Greenhills School's research program. As the host, I am responsible to engage and mentor senior high school students in the local K-12 community on a research project in my lab. I am also responsible to teach the students how to write scientific results and how to make a research poster.

Mechanical Engineering Graduate Council (President)

I am the President of Mechanical Engineering Graduate Council (MEGC) at U-M. As the president I am responsible for leading MEGC meetings, recruiting co-chairs, meeting with the graduate chair, and guiding the direction of MEGC. In the past, I have served as a co-chair of the mentoring, social, and DEI committees and have helped coordinate various department-level DEI activities, student mentoring activities, and social events.