Robots for law enforcement in India

The core argument

Technology has already reshaped how India banks, learns and governs. It is reasonable to ask whether policing should modernise too, and robots and AI clearly can help. But the honest version of that case is narrower and more demanding than "buy robots and roll them out." This paper argues one specific thing: India should adopt law-enforcement robots, but only in a bounded, accountable way, starting where the evidence of benefit is strongest and the risk to citizens' rights is lowest (hazardous tasks such as bomb disposal), under a clear legal framework and independent oversight, not as a shortcut to mass surveillance.

That distinction matters, because the most common way this technology is sold, city-wide cameras plus facial recognition, on the model of China, is exactly the version that has produced documented rights abuses and wrongful arrests. A robot that defuses a bomb keeps an officer alive. A face-recognition dragnet that misidentifies an innocent person does the opposite of justice. A serious modernisation plan has to tell those two things apart from the start.

1. Where robots genuinely help, and where the evidence is thin

1.1 The strong case: hazardous-task robots. The clearest, best-evidenced benefit of policing robots is keeping human beings away from things that explode. In Iraq and Afghanistan, roughly 2,000 iRobot 510 PackBots were deployed for explosive-ordnance disposal (EOD), reconnaissance and route clearance, letting bomb-disposal technicians inspect and neutralise devices from a safe standoff distance rather than approaching them in person (Army Technology). This is the model India should copy first: a robot doing a specific, dangerous, well-defined job that a human would otherwise do at mortal risk. India already handles bomb threats and suspicious-object call-outs; giving those squads better remote tools is a modernisation with an obvious, measurable payoff and almost no downside for ordinary citizens.

1.2 The weak case: general patrol and "RoboCop" surveillance. The evidence for general-purpose police robots is far thinner, and the original enthusiasm for humanoid patrol units does not survive contact with real deployments. The NYPD leased a Boston Dynamics "Digidog" and cancelled the contract in April 2021 after public backlash, having used it in only a handful of cases (Scientific American, 2023). Knightscope's patrol robots have been quietly dropped by police and municipal clients. These machines were marketed as cost-savers, but in practice they needed constant human minding, cost more than they saved, and generated more controversy than arrests. So the claim that robots reduce long-term costs by replacing salaries is not supported: policing budgets are indeed dominated by pay and pensions, but no force has actually cut headcount by buying robots, and the capital and maintenance costs are real. The saving comes from avoided harm in narrow high-risk tasks, not from replacing officers.

1.3 Claims to drop. Two arguments should simply be set aside because there is no evidence for them. The idea that a physically absent, robot-mediated police presence will reduce violence because "people not near police won't harm them" is speculative, it could equally increase resentment, as the Digidog backlash suggests. And the claim that AI "provides full transparency" and therefore reduces corruption reverses the actual record: as Section 3 shows, surveillance technology can entrench abuse as easily as it curbs it. Corruption in Indian policing is real, India sits in the lower-middle of Transparency International's Corruption Perceptions Index (91st of 180, score 39, in 2025), but the fix for corruption is accountability and audit, which are governance choices, not automatic properties of a camera.

2. The Dallas precedent: robots can take life, too

Before designing a framework, it is worth remembering what these machines can already do. In July 2016, after a gunman killed five officers in Dallas, police attached about a pound of C-4 explosive to a bomb-disposal robot and used it to kill the barricaded suspect. Robotics experts described it as the first time US police had used a robot for lethal force (PBS NewsHour, 2016). No charges followed; there was no rule written for it in advance.

The lesson is not that robots are evil. It is that a tool bought for one purpose (defusing bombs) was improvised into another (killing a person) with no policy, no oversight and no prior public debate. India should not import the technology and leave the rules to be improvised in a crisis. The framework has to come first.

3. The missing half: rights and accountability

Any plan to put robotic and AI systems into Indian policing has to answer three questions the original argument skipped: what happens to privacy, what happens when the technology is wrong, and who is responsible when a machine causes harm.

3.1 Privacy is a fundamental right, and surveillance must pass a test. In K.S. Puttaswamy v. Union of India (2017), a nine-judge bench of the Supreme Court held that privacy is a fundamental right under Article 21 (Human Dignity Trust, 2017). Privacy is not absolute, but any State intrusion must clear a proportionality standard: it must rest on a valid law, pursue a legitimate aim, and be no more intrusive than necessary. City-wide facial-recognition databases, the original paper's core "pre-induction" step, sit uneasily against that test. India's Digital Personal Data Protection Act, 2023 does now exist, but it grants the government broad exemptions for "security of the State," public order and crime prevention (Future of Privacy Forum, 2023), which means the discipline has to be built into the programme itself, not assumed from the law.

3.2 Facial recognition is measurably biased, and it is already producing wrongful arrests. The original held up facial recognition as a proven efficiency tool. The independent evidence says the opposite when it is used carelessly. The US National Institute of Standards and Technology tested 189 algorithms and found false-positive rates 10 to 100 times higher for Asian and African American faces than for white faces, and highest of all for African American women in one-to-many searches (NIST, 2019). The landmark "Gender Shades" study found commercial systems erred on up to 34.7% of darker-skinned women versus 0.8% or less of lighter-skinned men (Buolamwini & Gebru, 2018). This is not theoretical: the ACLU has documented at least 14 wrongful arrests in the US from facial-recognition matches, most of them Black people (ACLU, 2025), including a woman arrested while eight months pregnant.

India is not exempt. Under RTI, the Delhi Police confirmed it treats an 80% similarity score as a "positive match" and has run facial recognition against suspects from the 2020 Delhi riots, the 2021 Red Fort protest and the Jahangirpuri violence (Outlook India, 2022), that is, against people exercising the right to protest. At that same 80% threshold, an ACLU test found Amazon's tool falsely matched 28 members of the US Congress to arrest mugshots. A technology this error-prone cannot be the basis for detaining anyone; at most it can be an investigative lead requiring independent corroboration.

3.3 China is a warning, not a model. The original paper praised China's surveillance state and specifically named the contractor SenseTime as an example to follow. That example has since become one of the clearest cautionary tales in the field. China's facial-recognition and biometric surveillance is central to the documented oppression of Uyghurs in Xinjiang (Bulletin of the Atomic Scientists, 2022), and SenseTime itself was placed on the US Entity List (2019) and a US Treasury investment blacklist (December 2021) precisely because its facial-recognition technology was used in that surveillance (Wikipedia / US government designations). Building India's policing on that template would not be an efficiency win; it would be adopting the exact system that democracies are sanctioning.

3.4 Liability and use of force must be settled in advance. Autonomous and remote systems create an accountability gap: when a machine harms someone, "the robot did it" cannot be an answer. The framework must state, before any deployment, that (a) a named human officer authorises and is legally accountable for every action a policing robot takes, there is no autonomous use of force, ever; (b) existing use-of-force law applies to robot-mediated action exactly as it would to an officer's own hands, closing the Dallas-style gap; and (c) facial recognition, if used at all, is an investigative aid only and can never be the sole basis for an arrest.

4. A framework built around accountability

The original listed sensible technical prerequisites, better records, cybersecurity. Those still matter, but the ordering should change: rules and oversight come before hardware. A workable framework has four pillars.

• A legal basis and a proportionality gate. Each class of policing robot needs an authorising rule that names its permitted use and passes the Puttaswamy test. No general-purpose "surveillance infrastructure" mandate.
• An independent oversight and procurement body. Create a Police Technology Oversight Board, anchored at the existing Bureau of Police Research & Development (BPR&D) under the Ministry of Home Affairs but with independent (non-police) members, technologists, a data-protection expert, a civil-liberties representative, reporting publicly and to a parliamentary committee. It sets standards, approves each use case, and can halt deployments.
• Cybersecurity and data discipline. A weaponisable or hackable robot is a public danger; independent security certification is a precondition, not an afterthought. Any data collected is minimised, access-logged and time-limited.
• Transparency by default. A public register of every robot deployment and an independent audit of outcomes. Secrecy is what turned Delhi's FRT and the Dallas robot into problems.

5. A concrete near-term pilot (2026 to 2029)

Instead of a 35-year plan that lands "by 2050," India should run one tightly scoped, measurable pilot and let the results decide the next step.

• Bounded use case: explosive-ordnance disposal and hazardous-materials handling only. No surveillance, no facial recognition, no autonomous force. This is the use with real evidence (Section 1.1) and the smallest rights footprint.
• Who runs it: the BPR&D-anchored Oversight Board above, procuring to a common technical and safety standard and publishing the tender.
• Indicative cost: EOD robots run on the order of tens of lakhs to about ₹1 crore each (vendor list prices; to be tested by competitive tender). A starter fleet of roughly 10 units across selected metro bomb-disposal squads is on the order of ₹5 to 10 crore in capital plus training and maintenance, a small fraction of a single large state police force's annual budget, which runs into thousands of crore.
• Measurable success criteria (reviewed at 12 months, with a go/no-go gate):
  1. Share of suspicious-object / EOD call-outs handled with the technician at safe standoff distance (target ≥ 90%).
  2. Zero technician casualties in robot-handled call-outs.
  3. Mission logs, time to render safe, success/failure, recorded and independently audited.
  4. Any use outside the bomb-disposal mandate automatically triggers review and possible suspension.
• Scale only on evidence: expansion to a second task or city happens only if the pilot meets its criteria and clears an independent audit, not on a fixed calendar.

6. Conclusion

The original instinct is right: Indian policing should modernise, and robots have a real place in it. But the version worth building is disciplined. Start with the job robots demonstrably do well and safely, taking human beings away from bombs, under a named oversight body, clear liability rules and public audit. Refuse the mass-surveillance template that has produced wrongful arrests abroad and sanctions against its own vendors. Prove the benefit on a small, measured pilot before scaling. Done this way, robots make policing safer without making citizens less free. Done the other way, they simply automate old abuses at new speed.

Sources

1. Human Dignity Trust (2017): Puttaswamy v. Union of India, privacy held a fundamental right under Article 21 (nine-judge bench).
2. Future of Privacy Forum (2023): India's Digital Personal Data Protection Act, 2023, and its broad State exemptions.
3. NIST (2019), FRVT Part 3 / NISTIR 8280: false positives 10 to 100× higher for Asian and African American faces than white faces.
4. MIT News (2018): Buolamwini & Gebru, "Gender Shades", up to 34.7% error for darker-skinned women vs ≤0.8% for lighter-skinned men.
5. ACLU (2025): at least 14 documented US wrongful arrests from facial recognition, most of them Black people.
6. Outlook India (2022): Delhi Police treats 80% similarity as a "positive match"; FRT used on riot/protest suspects; Amazon tool falsely matched 28 US Congress members.
7. SenseTime: US Entity List (2019) and Treasury blacklist (Dec 2021) over facial-recognition surveillance of Uyghurs.
8. Bulletin of the Atomic Scientists (2022): China's facial-recognition/biometric surveillance drives the oppression of Uyghurs.
9. PBS NewsHour (2016): Dallas police used a bomb-disposal robot with C-4 to kill a sniper, first police use of a robot for lethal force.
10. Army Technology: ~2,000 iRobot 510 PackBots deployed in Iraq/Afghanistan for EOD and reconnaissance at safe standoff distance.
11. Scientific American (2023): NYPD terminated its Digidog lease in April 2021 after public backlash.
12. Transparency International CPI: India 91st of 180 (score 39) in 2025, persistent public-sector corruption.