Interview
Vijay Kumar: Flying Robots | Lex Fridman Podcast #37
- By 2030, the price-to-performance ratio of robotic components is predicted to drop significantly, rendering individual unit costs negligible.
- Mature sensors and computers are expected to make small unmanned aerial vehicles standard for creating and deforming three-dimensional midair patterns.
- Robotics will transition toward data-driven and learning-based approaches that operate alongside traditional model-based methods to handle unmodeled aerodynamic effects.
- Relying solely on computer vision and machine learning for safety-critical applications is expected to saturate due to exponentially increasing data and power requirements for accuracy improvements from 99% to 99.9%.
- Short-distance flying cars are predicted to rely on fossil fuel engines rather than electric power due to fundamental limits in energy and power density.
- Significant breakthroughs in battery technology are necessary prerequisites for the economic and physical viability of large-scale electric flying vehicle deployment.
- True autonomy, defined as navigation without GPS, communications, or prior maps, remains a challenging goal requiring the mitigation of current infrastructure brittleness.
- Human-robot collaboration is expected to evolve toward robots modeling human state and attention, such as monitoring eyes and circadian rhythms, to enhance safety.
- A significant risk exists regarding the ease of weaponizing autonomous swarms, where a single agent's success can achieve a destructive mission.
- Technological governance requires politicians to become literate in technology and engineers to be elected to office.
- Current Level 5 self-driving capabilities confined to parking lots are expected to struggle in complex, unstructured environments like the streets of Mumbai or Naples.
- Predicting the future is advised as a skill for students to sharpen their senses and become smarter, despite the frequent inaccuracy of specific predictions.
- A strong foundation in mathematics and representations is required for future robotics, as deep neural networks alone cannot resolve issues of explainable AI.
- The timeline for flying cars implies that hybrid or fossil-fuel solutions will likely precede fully electric autonomous flight due to challenges in the "clean" aspect versus the feasibility of the "autonomy" aspect.