LiveNet: Robust, Minimally Invasive Multi-Robot Control for Safe and Live Navigation in Constrained Environments; From Production Challenges to Logistics Solutions: A Multi-Layered Analysis of Autonomous Vehicle Integration Using the Tesla Manufacturing Case

Sociotechnical Synthesis
(Executive Summary)
Safe and Effective Autonomy
“We had too many robots… we had to get rid of them all”
- Elon Musk, CBS Mornings, 2018

Robotics provides an interface for computers to manipulate and interact with the physical world. This powerful capability has the potential to provide abundance and increase humanity’s prosperity through its efficiency and economic growth. The overarching goal of my research is to accelerate the development of this STS regime by providing safe solutions and guidance. My technical research extends existing robotics control methods to resolve deadlock scenarios between multiple agents in a safe and lively manner. My STS research was a higher-level analysis of Tesla’s 2018 autonomy induced manufacturing failure – a period termed as “production hell” – to provide guidance on safely developing and deploying autonomous vehicles in the logistics sector.
My technical project, conducted under the guidance of Professor Rohan Chandra, sought to employ liveness constraints in robotic controllers to solve deadlocks between agents in symmetric situations. Deadlocks occur when the ideal path of two or more agents intersect and environmental symmetry is when kinodynamic properties of all the agents are the same – the robots’ position, velocity, acceleration all perfectly mirror one another. This situation, termed social mini games (SMG), is shown in figure 1 where both robots must past through a narrow gap to get to their goals in a symmetric environment. Our solution was to adopt an existing neural control method and add liveness constraints (a set of rules to solve SMG scenarios) in the controller’s control barrier function (CBF) to allow for safe deadlock resolution. The safe trajectories our custom neural control method, termed LiveNet, generates is shown in figure 2.
Social Mini Game Depiction

Figure 1: This figure depicts a social mini game where both robotic agents must pass through a narrow doorway to reach their goal.
LiveNet Social Mini Game Solution

Figure 2: Depicts LiveNet solving the doorway social mini game by speeding up the red agent and slowing down the blue agent to avoid a conflict.

My STS research made use of the multi-layered perspective (MLP), outlined by Frank W. Geels in his 2007 analysis of the Dutch highway system, to analyze how autonomy resulted in Tesla’s 2018 “production hell” and use these lessons to provide guidance in the development and deployment of autonomous vehicles for logistics delivery purposes. The MLP breaks socio-technical systems (STS) into three layers: the niche (where technological advancements take place), the regime (the mainstream system supported by traditional social norms), and the landscape (where global trends occur). My MLP research of Tesla’s production hell period showcased a concerning interaction where niche developments are picked up by the landscape reinforcing global trends which ultimately pressure the regime to adopt the niche development – regardless of the development’s merit in the regime. For example, advancements in robotics paired with Tesla’s customers’ and shareholders’ belief that the company should strive to be the company of the future caused Tesla to replace traditional automotive assembly lines with highly automated processes resulting in manufacturing failures. To prevent the negatives of this MLP interaction in the case of logistics and delivery, I advise the importance of the active participation of potential displaced actors (such as vehicle drivers) in the design and development of autonomous vehicles. Drawing on their first-hand experience, these drivers and operators can readily identify and provide edge case scenarios that may cause methods previously considered robust to fail.
My research across both the technical and STS portions of my capstone allowed me to expand my perspective beyond the narrow view of the discourse of inevitability. While I do believe the growing role of artificial intelligence and robotics in society is inevitable, I now see we can actively shape its growth to ensure we maximize its positive externalities while minimizing its negative ones. The key in this revelation is the shaping is active and as engineers, we must do our best to ensure products, services, and solutions we engineer are designed and deployed in a responsible manner. To take from my STS research, engineers should consider actors impacted by their solutions and specifically actively identify and involve actors that are displaced by their solutions.
I would like to acknowledge Professor Kathryn A. Neeley for her significant contributions for my understanding of socio-technical systems and her guidance throughout my STS research. As aforementioned, I would like to acknowledge Professor Rohan Chandra for his guidance and active involvement with my technical research.

 
References
Chandra, R., Zinage, V., Bakolas, E., Biswas, J., & Stone, P. (2023). Decentralized multi-robot social navigation in constrained environments via game-theoretic control barrier functions. arXiv preprint arXiv:2308.10966.

Geels, F. W. (2007). Transformations of Large Technical Systems: A Multilevel Analysis of the Dutch Highway System (1950-2000). Science, Technology, & Human Values, 32(2), 123-149. https://doi.org/10.1177/0162243906293883

King, G. (2018, April 13). Elon Musk: “Tesla CEO Elon Musk offers rare look inside Model 3 factory.” CBS Mornings. https://www.youtube.com/watch?v=lOz7cPJQd8E

School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisor: Rohan Chandra
STS Advisor: Kathryn Neeley
Technical Team Members: Srikar Gouru