The Future of Robot Perception: Kimera-Multi and Advancements in AI
Understanding the world around us is a complex task that humans effortlessly accomplish every day. Now, MIT Laboratory for Information and Decision Systems (LIDS) researchers Luca Carlone and Jonathan How are asking the question: Can robots perceive their environment in a similar way? They believe they can, and they have developed an open-source library called Kimera that enables robots to construct real-time 3D maps and identify objects. Their latest project, Kimera-Multi, takes this technology further by allowing multiple robots to communicate and create a unified map. The remarkable achievements of their research were recognized with the prestigious IEEE Transactions on Robotics King-Sun Fu Memorial Best Paper Award for 2022.
Advantages of Scaling the System
By increasing the number of robots working together to generate 3D maps, there are several potential advantages. Consistency is one key benefit – each robot can create its own map that is self-consistent, but by collaborating, the team can create a globally consistent map of the world. This allows for more accurate and comprehensive mapping. Additionally, having multiple robots increases redundancy, which is especially important in scenarios like search-and-rescue missions. If one robot encounters an issue, there are others that can continue the task and increase the chances of success. Scaling up the team of robots also means tasks can be completed more quickly.
The Lessons Learned and Challenges Overcome
In recent experiments, Carlone and How conducted a mapping experiment on the MIT campus involving eight robots. These robots had no prior knowledge of the campus or GPS assistance. Their task was to estimate their own trajectory and build a map of the environment. The challenge was to make the robots understand the environment as humans do – not just the shape of obstacles, but also the semantic information about objects, such as chairs and desks. The robots exchanged limited information to enhance their maps when they met, utilizing a distributed protocol that focused on exchanging specific 3D coordinates and clues extracted from sensor data. The goal is to move from capturing basic semantics to a more hierarchical representation of the environment.
Future Applications of Kimera and Similar Technologies
Kimera and similar technologies have the potential for various applications. In the field of autonomous vehicles, the ability to communicate and share information between vehicles could significantly improve mapping and model building. This would lead to improved safety and access to data from multiple perspectives. In search and rescue scenarios, Kimera could be used to explore forests and map buildings after natural disasters, assisting first responders in locating survivors. Additionally, in future flexible factories, robots equipped with Kimera-like capabilities could work alongside humans in less structured environments, increasing efficiency and adaptability.