Abstract
Shared autonomy in self-driving vehicles refers to a collaborative control model where both humans and autonomous systems work together to make decisions in real-time. This concept is crucial for the safe deployment of autonomous vehicles (AVs), as it blends the strengths of both human intuition and machine precision.
One of the primary reasons shared autonomy is vital for self-driving cars is safety. Although autonomous systems are designed to reduce human error, they are not infallible. Situations may arise where the vehicle's sensors or algorithms struggle to interpret complex conditions, such as unpredictable weather, erratic behavior from other road users, or road construction zones. In such cases, human intervention becomes necessary. This transition between human and machine control should be seamless and intuitive to avoid confusion or delay during critical moments.
Moreover, shared autonomy fosters trust between humans and autonomous systems. For passengers who are new to self-driving technology, knowing they can take control if needed provides a sense of security. This trust is essential for the adoption of AVs, as people may be hesitant to relinquish control entirely to a machine. As autonomous systems continue to improve, the level of human intervention required may decrease, but in the interim, shared autonomy allows for a balanced approach that builds confidence in the technology while ensuring safety. It is a key factor in encouraging both regulatory bodies and the public to embrace autonomous vehicles on a large scale.
Additionally, shared autonomy is crucial for the interaction between self-driving cars and non-autonomous vehicles. Self-driving cars must be able to communicate with human drivers and adapt to their behavior. For example, the vehicle’s system may recognize that a human driver is acting unpredictably or entering a dangerous situation, allowing the car to adjust its driving behavior to mitigate potential risks. Shared autonomy helps establish this kind of adaptive interaction, ensuring a smoother transition as the roadways evolve to accommodate both human drivers and autonomous technology.
For this project , we propose to leverage an immersive driving simulator, developed with the support of previous Safety21 projects, to expose a cohort of 20 drivers to shared autonomy. We will recruit a cohort of 10 young drivers, ages 25 to 30, who are more likely to embrace self-driving, and a second cohort of 10 more mature drivers, aged 40 to 50 who might be more skeptical of self-driving. We will expose them to a 15 minute challenging scenario so they experience manual, full self-driving and a hybrid model where a Guardian lets them drive manually while being supervised by the automated agent. We will collect survey data as well as simulator data. Shared autonomy is a cornerstone for the development of self-driving cars, combining the reliability of autonomous systems with the experience and judgment of human drivers. By ensuring safety, fostering trust, and enabling effective interactions between autonomous and non-autonomous vehicles, shared autonomy paves the way for a more integrated, efficient, and secure future of transportation.
Description
Timeline
Strategic Description / RD&T
Section left blank until USDOT’s new priorities and RD&T strategic goals are available in Spring 2026.
Deployment Plan
The proposed work will involve the following key milestones:
Q1) For the first quarter, we will develop driver user interfaces and connect them with Autoware to share information between the vehicle state and the driver state. We will implement our Local Control Barrier Functions for the Safety Control of Hybrid Systems into the Metodrive driving simulator developed by Jitsik LLC.
Q2) For the second quarter, we will develop and test a 15 mn drive scenario with urban and highway components, that offers challenges for manual and automated driving. We will also develop an IRB with the University of Pennsylvania to test with a cohort of 20 drivers.
Q3) For the third quarter, we will run a set 10 young drivers, ages 25 to 30 and a second test of drivers age 40 to 50, and assess their use and level of comfort with Shared Autonomy. The Metodrive VR simulator will be used.
Q4) For the fourth quarter, we will analyze the data collected, inform the data management, prepare for publication and final report.
Expected Outcomes/Impacts
Shared control in autonomous driving refers to a system where both the vehicle’s autonomous system and the human driver collaborate in decision-making and control. This approach offers significant benefits, enhancing both safety and user experience. One key benefit is increased safety. In situations where the autonomous system encounters uncertainty—such as complex traffic or adverse weather conditions—the human driver can intervene if necessary, ensuring that decisions align with human judgment and intuition. This helps reduce the risk of accidents that could arise from an over-reliance on automation, particularly in unpredictable or ambiguous scenarios. Shared control also improves driver comfort and trust. By maintaining an active role in the driving process, even if limited, the human driver feels more in control, which can alleviate anxiety or frustration with fully autonomous systems. This sense of partnership between the vehicle and the driver fosters trust in the technology, encouraging more widespread acceptance of self-driving cars. We anticipate the use of Shared Control will make it easier for stakeholders to develop legislation to accommodate self-driving in all future vehicles.
Expected Outputs
The proposed work’s main anticipated outputs include:
- Research Impact: The proposed work will develop Guardian Angel techniques that will allow drivers to drive “manually” while benefiting from the expertise of an Automated Driving Agent. The perspective of blending manual and automated driving is key to the wide adoption of self-driving, which is seen as one of the best ways to improve safety on our roads..
- Publications: Research developed through the proposed work will be disseminated through publications at major international venues such as IEEE, TRB Annual Conference, TRR Transportation Journal, SAE, and ITS.
- Software: Source code and associated software systems will be developed.
- Student Training: Master and undergraduate students will be trained during the course of the research.
TRID
The proposal uniqueness stems from our use of a novel algorithm for shared autonomy. We propose the use of Local Control Barrier Functions for the Safety Control of Hybrid Systems. Our TRiD search using the project keywords resulted in 16 entries (see uploaded file), which range from 2006 to 2024. Half of the results refer to work done in the last three years which testifies to the novelty of the research. Two of the references (1) and (5) refer to our previous research through the Safety21 UTC. If we exclude the out of context references such as (2) which relates to air transport, (16) which relates to telerobotics, (13) which relates to the creation of 3D environments, (6) which relates to shared roads, only a handful of the remaining references are relevant to our work. (3) and (10) actually human behavior in the context of autonomous vehicles. (7) is an interesting review on human-centered collaborative automated driving which we will integrate in our research, as well as (8) which offers a shared control scheme for semi-autonomous driving.
Individuals Involved
| Email |
Name |
Affiliation |
Role |
Position |
| erickg@upenn.edu |
Guerra, Erick |
University of Pennsylvania |
PI |
Other |
| LoebH@email.chop.edu |
Loeb, Helen |
Jitsik LLC |
Co-PI |
Other |
| rahulm@seas.upenn.edu |
Mangharam, Rahul |
University of Pennsylvania |
Co-PI |
Other |
Budget
Amount of UTC Funds Awarded
$35000.00
Total Project Budget (from all funding sources)
$70000.00
Documents
Match Sources
No match sources!
Partners
| Name |
Type |
| Jitsik LLC |
Deployment Partner_ Deployment Partner_ |
| The Autoware Foundation |
Deployment Partner Deployment Partner |
