Overview and useful information for the public
As autonomous robots become increasingly capable of learning, adapting and operating across different environments, questions about how these systems should be designed, deployed and governed become increasingly important.
PILLAR-Robots has explored these questions alongside its technical research into purposeful, lifelong and open-ended autonomy. The project’s new Policy Brief brings together selected recommendations developed through its socio-economic, ethical, legal and technological work, translating research findings into practical guidance for the wider AI, data and robotics ecosystem.
The document focuses particularly on autonomous robotics, including general-purpose systems able to operate across different domains and robots with open-ended learning capabilities that may continue adapting after deployment.
Rather than treating regulation, ethics, technology and socio-economic impact as separate issues, the Policy Brief examines how these dimensions interact as robots move from controlled research settings into real-world applications.
Six topics, one connected framework
The Policy Brief is organised around six interconnected policy topics, moving from questions about what autonomous robots should be able to do to what should happen when their actions cause harm.
The six areas are:
- Context-dependent roles and learning capabilities of autonomous robots
- Alignment under uncertainty in autonomous robots
- Meaningful human oversight of autonomous robots
- Equitable and sustainable access to autonomous robots
- Safety, autonomy and privacy of direct and indirect users
- Legal responsibility and financial security for robot-caused harm
Together, they provide a framework for examining autonomous robotics not simply as a technological challenge, but as a combination of technical capabilities, deployment choices, societal expectations and regulatory responsibilities.
Matching autonomy and learning to the context
One of the central findings of the Policy Brief is that there is no single appropriate level of autonomy for every robot or application.
A robot may operate as an assistive tool in one setting, as a collaborative system in another, or perform functions with a greater degree of independence elsewhere. The appropriate role depends on factors such as the task, operational environment, technical capabilities and potential risks.
The same principle applies to open-ended learning.
While the ability to continuously acquire new skills can make robots more adaptable, unrestricted learning may not be appropriate in safety-critical, highly regulated or rights-sensitive environments.
The Policy Brief therefore recommends that robot autonomy and learning capabilities should be adaptable to the context and proportionate to the risks involved. It also calls for further research and standardisation around concepts such as general-purpose robotics, open-ended learning and adaptive autonomy.
The impact on work is another part of this discussion. Autonomous robots can change job content, task allocation and working conditions, making worker representation, training, reskilling and appropriate workplace governance important elements of responsible deployment.
Alignment as an ongoing process
For increasingly autonomous systems, alignment cannot necessarily be considered complete when a robot leaves the laboratory.
Real-world environments contain situations, interactions and uncertainties that cannot always be reproduced during development and testing. As a result, the Policy Brief approaches alignment as an ongoing, context-sensitive process.
Among the recommendations are the use of controlled pilots, staged deployment and continuous monitoring before progressively expanding a system’s scope or autonomy.
Regulatory and experimental sandboxes can also provide environments where the technical, organisational, ethical, environmental and societal consequences of autonomous systems can be assessed before widespread deployment.
Transparency is equally important, but the report does not treat it as an all-or-nothing requirement. Information should instead be presented in ways that help users understand what a robot can and cannot do without creating unnecessary complexity, cognitive overload or security risks.
Making human oversight meaningful
Human oversight is often proposed as a safeguard for autonomous AI and robotics, but its presence alone does not guarantee effective control.
According to the Policy Brief, meaningful oversight depends on whether the people responsible for supervision actually have the information, time, authority, organisational conditions and technical means necessary to intervene effectively.
This includes defining safe fallback behaviours for unfamiliar situations, maintaining dependable human override mechanisms and ensuring that oversight does not disappear simply because a system has demonstrated increasing reliability.
The report also considers practical human factors such as automation bias and mental overload. Asking one person to supervise too many robots, alerts or decisions can undermine the effectiveness of oversight itself.
Different environments may therefore require different combinations of real-time monitoring, periodic review, escalation procedures, audit logs and post-deployment evaluation.
Importantly, the Policy Brief also distinguishes between direct users, who deliberately adopt or operate a robot, and indirect users, who may be affected by it without having participated in the decision to deploy it. These indirect users should also be considered in risk assessment and oversight frameworks.
Equitable and sustainable access to autonomous robotics
The benefits of advanced robotics will depend partly on who can realistically access and deploy these technologies.
SMEs, cooperatives, rural actors, public organisations and organisations with limited technical resources may face barriers including acquisition costs, maintenance requirements, infrastructure limitations, regulatory complexity and a lack of specialist expertise.
The Policy Brief therefore connects robotics innovation with the broader question of equitable access.
Recommendations include encouraging modular and adaptable robotic systems, supporting interoperability through standards and open protocols, and ensuring that regulatory requirements are proportionate and accompanied by practical guidance and technical assistance.
Access also depends on information. Users must be able to understand what robotic systems do, their limitations and the risks and benefits associated with their use.
The report consequently highlights robotics literacy, accessible information and long-term environmental and territorial sustainability as part of the wider conditions for meaningful access.
Protecting users: safety, autonomy and privacy
As robots increasingly operate in workplaces, healthcare, education and shared environments, safety extends beyond preventing physical collisions or hardware failures.
The Policy Brief identifies broader risks related to unpredictable behaviour, cybersecurity vulnerabilities, overreliance, manipulation and inappropriate forms of emotional attachment.
User autonomy is particularly relevant where people may have limited ability to avoid robotic interaction.
Among the recommendations is maintaining a human alternative in essential or rights-affecting services where meaningful human contact is necessary for dignity, care, trust or the exercise of users’ rights.
The document also calls for design approaches that avoid coercive or manipulative interactions, including deceptive interfaces, misleading anthropomorphic cues or mechanisms that make it unnecessarily difficult for users to refuse or disengage from interaction.
Privacy presents another challenge because autonomous robots can collect information through cameras, microphones, sensors, biometric systems and interaction histories while moving through physical and social environments.
This means data collection may affect not only direct users but also workers, visitors, patients, family members and bystanders.
The Policy Brief therefore recommends data minimisation and layered, context-appropriate privacy information, together with visible or auditory indications when robots are collecting and processing personal data in shared environments.
Responsibility when autonomous robots cause harm
As robotic systems become more complex, responsibility may be distributed across designers, developers, providers, deployers, operators, maintainers and other actors.
This can make it difficult to determine who should be responsible when something goes wrong.
The Policy Brief argues that responsibility should primarily remain with the human actor best positioned to control, prevent or mitigate the relevant risk, depending on whether the problem originates from design choices, software updates, deployment conditions, maintenance, misuse or operational decisions.
It also recommends avoiding the direct attribution of criminal liability to robots themselves.
At a broader European level, differences between national civil liability regimes may create uncertainty for victims, companies and innovators. The report therefore calls for liability approaches that account for risk, actor capacity and context while protecting victims without imposing disproportionate burdens on low-risk applications or smaller organisations.
Similarly, financial security requirements such as mandatory insurance should not automatically apply to every autonomous robot. Their necessity should depend on the technology, deployment context and potential severity of harm.
From interdisciplinary research to practical recommendations
The Policy Brief is based on work carried out within Work Package 10 of PILLAR-Robots, combining socio-economic research, legal analysis, ethical assessment and interactive ethics workshops involving project developers.
This interdisciplinary approach is particularly relevant for autonomous robotics because issues such as safety, autonomy, transparency, oversight, access and responsibility cannot be addressed independently.
Technical design decisions influence legal responsibilities. Regulatory requirements can influence access to innovation. Oversight mechanisms affect both safety and working conditions. Data collection can enable robotic capabilities while simultaneously creating privacy risks.
The recommendations therefore target a broad range of actors, including policymakers, regulators, standardisation bodies, developers, providers, employers, deployers, researchers and civil society organisations.
Looking beyond the project
As PILLAR-Robots approaches the conclusion of its research activities, the Policy Brief provides a bridge between the technologies investigated within the project and the wider discussion about how increasingly autonomous robots should enter society.
Its central message is not that autonomous robotics requires a single fixed governance model. Instead, governance needs to remain sensitive to context, risk, users and the evolving capabilities of robotic systems.
Autonomy can change. Robots can continue learning. Deployment environments differ. New risks may emerge after systems leave the laboratory.
For these reasons, the Policy Brief points towards governance based on proportionality, gradual deployment, meaningful oversight, stakeholder participation and clear allocation of responsibility.
The complete Policy Brief contains the selected findings and recommendations across all six policy areas and provides references to the wider set of recommendations developed within PILLAR-Robots.