Hyderabad-based startup Perceptyne is building semi-humanoid robots that “are as flexible and adaptable as humans” for industrial settings.
Ever stop and wonder how easy it is to pick up an object placed anywhere on a table and change its position and orientation without solving a trigonometric equation in a split second? Perceptive tasks, which come quickly to humans, are complex for robots. Hyderabad-based startup Perceptyne is trying to build an automation system that is as flexible and adaptable as a human.
EFY spoke to co-founder (CBO and co-CTO) Mrutyunjaya Nadiminti, who explained that automating complex assembly tasks requires integrating multiple technologies: robotic arms, grippers, image processing systems, and more. “Manufacturing companies either build an in-house team for this or rely on system integrators. However, every project is unique for system integrators, requiring multiple iterations. If a component does not work, it might need to be replaced by an overseas vendor, making the process lengthy. Many automation projects are thus abandoned mid-way due to long integration cycles and costs,” he elaborates.
To resolve this issue, Mrutyunjaya along with his two co-founders Raviteja Chivukula (CEO and co-CTO) and Jagga Raju N (COO) got together in 2022 to develop an AI-powered perception system for industrial settings that aims to understand the environment like a human, with a strong intelligence layer for decision-making, and an actuator system capable of human-like dexterity.
Humanising the Bot
Mrutyunjaya contests that while humans use a combination of vision, touch, and force sensing to perform tasks intuitively, why can’t robots do the same? Perceptyne’s PR-34D is equipped with multimodal tactile, force and vision sensors and actuators that allow it to gather better perception of the placement or pose of an object, obstacles in the environment etc. as compared to traditional robots which rely mainly on position sensors, or an integrated vision-based sensing system in newer models expanding capabilities to a limited extent.
However, rich sensing can only be useful if the robot is flexible enough to maneuver in multiple directions and into narrow spaces. Traditional robotics states that the more degrees of freedom (DoF) the robot has, the more flexible and skilled it is. PR-34D comes with two 7-DoF arms, with 3-DoF at the shoulder, 1-DoF at the elbow and 3-DoF at the wrist, each attached to a 10-DoF ‘three-fingered hand’ which allows the robot to navigate complex spaces, pick up objects in different orientations, and manipulate them with precision. Depending on the need, the startup has also used traditional two-finger and vacuum grippers to handle objects.
A 34-DoF robot has infinite ways to reach a target position that introduces redundancy, which can be an advantage but also increases computational complexity. Mrutyunjaya explains, “Our AI-driven approach optimises this redundancy, allowing the robot to select the most efficient movement path based on the task, workspace constraints, and real-time sensor feedback. This ensures smooth, collision-free motion while maximising efficiency.”
The robot’s multi-stage hierarchical AI architecture, currently in the domain of neuro-symbolic AI, “mimics human thought processes, enabling the robot to reason and adapt dynamically rather than following rigid pre-programmed instructions,” says Mrutyunjaya.
Unlike traditional AI models that rely solely on deep learning, integrating machine learning with symbolic logic allows the robot to understand cause-and-effect relationships, make predictions, and adjust its movements as a human would when learning a new task. “We are using object detection and pose estimation models which process inputs from RGB and depth-sensing 3D cameras to estimate an object’s position and orientation so that the robot can plan its motion. With path prediction models, we help the robot navigate complex spaces to avoid obstacles. We focus on teleoperative training models. A human operator demonstrates a task, and the robot records sensory data during these demonstrations,” elucidates Mrutyunjaya.
The PR-34D is not designed for heavy lifting but for tasks that require precision and dexterity—assembling components, performing quality checks, and handling intricate objects. “If you think of a human at a workstation performing a repetitive, fine-motor task, that is exactly what our robot is designed to automate,” Mrutyunjaya clarifies.
Moving beyond prototypes
Perceptyne has developed its core technology in-house, from mechanical design and electronics to embedded control systems and AI models. Currently, the startup is at the MVP stage, collaborating with T-Hub and T-Works in Hyderabad for prototyping and manufacturing support. Initially incubated at IIIT Hyderabad, the startup employs around 25 people and actively seeks to expand its R&D team.
Perceptyne is building its Version 4 prototype in-house for internal testing and pilot deployments. However, as production scales to tens and hundreds of units, they plan to expand their vendor ecosystem in India, capitalizing on the “Make in India” initiative while keeping final assembly under their control.
Mrutyunjaya says that Perceptyne’s robots can be deployed in automotive and electronics manufacturing, with pilot projects already underway with some of the world’s largest multinational corporations (MNCs). While initial deployments are happening in India, the startup is eyeing the US and Europe as its principal target markets, where labour shortages and reshoring trends drive demand for automation.
“Our goal is not just to build a robot but to redefine automation,” says Mrutyunjaya. “By making industrial robots as adaptable and intuitive as humans, we eliminate the barriers that have kept automation from reaching its full potential.
