Talk like a roboticist

Coopetition — engaging in both cooperation and competition to improve performance and outcomes

Cross-capability multi-robot adaptation — the ability of a group of robots with different skills and specialisations to work together, solve problems and adapt to new situations

Fault tolerance — a system’s ability to continue operating effectively when some of its parts fail

Introspection — the ability to observe and examine one’s own thought processes and motivations

Lifelong autonomy — the ability of a robot to continually adapt and learn throughout its lifetime, without input from humans

Robots provide many services for us, from delivering packages to exploring other planets; however, many tasks are too complex for a single robot to handle alone. Teams of robots, especially those with complementary abilities, could tackle much bigger and more difficult challenges in a range of situations and environments.

For example, in the aftermath of a natural disaster, such as an earthquake or a tsunami, teams of robots could search areas that are too dangerous for humans to enter. “Drones can scout an area from above, ground robots can navigate through debris or rough terrain, and specialised robots can manipulate objects or access confined spaces,” explains Dr Hao Zhang from the University of Massachusetts Amherst. “Robot teams can offer efficient, robust and adaptive solutions to societal challenges, not just in disaster response but also in space exploration, agriculture, manufacturing and infrastructure inspection.”

Lifelong autonomy

For robot teams to be truly useful in the real world, they need to be made up of robots that can learn from experience, adapt to new situations and improve their performance over time without human input – a concept known as lifelong autonomy. “This is essential for long-term deployment outside of controlled settings, enabling robots to handle unforeseen situations, recover from failures, and evolve alongside changing tasks and teammates,” explains Hao.

Developing lifelong autonomy is difficult and involves creating robots that can accurately perceive the world around them, reflect on and assess their own performance, and continually learn and adapt to new situations. Robots must have these abilities of perception, introspection and adaptation if they are to work effectively as part of a team.

Assembling a team

For robot teams to succeed, they must coordinate a range of different abilities, communicate effectively and make joint decisions – all while avoiding errors and working safely. The robots must adapt to the evolving behaviour of their teammates in real-world environments where conditions can change quickly and robots can malfunction. This level of problem solving and improvisation is difficult enough to achieve in single robots, let alone teams of robots with a range of skills and capabilities.

To develop successful robot teams, Hao is drawing on key insights from social psychology and human teamwork. “Humans are the most adaptable species on Earth,” says Hao. “Collective intelligence, emerging from collaboration, shared effort and internal competition, allows human teams to adapt swiftly to unexpected changes in their environment or composition.” Whether in sports teams, multi-national companies or academic research groups, functional heterogeneity – the useful range of skills and abilities within a team – enables people to tackle complex tasks effectively.

In contrast, robot teams lack this level of adaptability. “One project that we are working on, called Autonomous Group Introspective Learning and coopEtition (AGILE) for Cross-Capability Multi-Robot Adaptation, aims to enable diverse robot teams to respond effectively to novel situations and complex failures,” says Hao.

Group introspection

One of the key goals of AGILE is to help robots reflect on how well they are working together as a team. “Autonomous group introspection refers to the ability of a team of robots to collectively monitor, analyse and understand their own behaviours, roles and performances without human intervention,” explains Hao.

By sharing what they see, what they are doing and how their tasks are progressing, robots can collectively figure out if something is not working, adjust their roles and come up with new plans. For example, if one robot malfunctions, group introspection would allow the team to replace the broken robot with another teammate that has similar capabilities. This kind of shared awareness enables resilient, flexible and efficient teamwork, allowing robot teams to work together effectively on complex tasks in the real world.

Coopetition

As strange as it may seem, a healthy level of competition within a team can improve teamwork. “Cooperative competition – or coopetition – allows robots to collaborate toward a shared goal while simultaneously competing to pursue individual objectives and optimise their own performance,” explains Hao. Cooperation allows the team to tackle objectives that are too complex for individual robots to solve by themselves, while competition encourages each robot to continually improve their performance. “This balance allows the robot team to efficiently allocate tasks to individual team members when solving complex problems, enhances its resilience to individual failures and improves its adaptability in evolving environments,” continues Hao.

Looking to the future

Currently, AGILE focuses on developing teams of robots that can collectively make decisions and coordinate their actions effectively. But the project’s next steps aim even higher. Hao hopes to develop collaborative perception, enabling robots to use their own sensors to monitor the health and abilities of their teammates in real time. The AGILE project will also explore human-robot teaming, allowing people and robots to work together safely by combining their unique strengths. “If successful, AGILE will mark a critical step toward building resilient robot teams capable of near human-level adaptability, transforming multi-robot applications through greater fault tolerance, agility and adaptability,” says Hao.

Dr Hao Zhang
Associate Professor of Computer Science, Director of the Human-Centered Robotics Laboratory, University of Massachusetts Amherst, USA

Field of research: Robotics

Research project: Autonomous Group Introspective Learning and coopEtition (AGILE) for Cross-Capability Multi-Robot Adaptation

Funders: US Defense Advanced Research Projects Agency (DARPA); National Science Foundation (NSF)

Website: https://hcr.cs.umass.edu

About robotics

“Robotics is an exciting field at the intersection of engineering, computer science and artificial intelligence (AI), with the potential to transform nearly every aspect of society,” says Hao. “It is a field where cutting-edge innovation directly shapes real-world impact.” As robots become more intelligent and autonomous, the next generation of roboticists and engineers will play a crucial role in shaping how they are developed and used. Career opportunities in robotics range from designing robotic hardware and developing AI software to working in research labs, startups, industry or government.

However, robotics is not without its challenges. “One of the most challenging aspects of robotics is mastering its interdisciplinary nature,” says Hao. “Another major challenge lies in handling the complexity and unpredictability of real-world environments.” The real world is full of obstacles, noise and changing conditions, and is far more complex than controlled lab environments or computer simulations. “Overcoming these challenges requires deep expertise in a core discipline, such as AI for robot adaptation, alongside the ability to synthesise knowledge across fields and build integrated systems through interdisciplinary collaboration,” explains Hao.