CtrlK
| TALOS |
|---|
 |
| General information |
| Manufacturer |
| Country of origin |
| Year unveiled |
| Status |
| Type |
| Purpose |
| Predecessor |
| Successor |
| Price |
| Website |
TALOS is a full-size, torque-controlled humanoid robot developed by PAL Robotics, a robotics company headquartered in Barcelona, Spain. Unveiled in 2017, TALOS was designed from the ground up as a high-performance research platform targeted at industrial applications, with the long-term goal of having humanoid robots work side by side with humans in factory environments. The robot's name references the mythological bronze automaton that protected the island of Crete in Greek legend.
Standing 175 cm tall and weighing approximately 95 to 100 kg (depending on configuration), TALOS features 32 degrees of freedom, torque sensors in all actuated joints, and an EtherCAT communication bus that enables control loops running at up to 5 kHz. Each arm can carry a 6 kg payload at full extension, making the robot suitable for handling heavy industrial tools. TALOS runs entirely on ROS (Robot Operating System) and ships with open-source simulation support through Gazebo. As of 2020, at least six TALOS units had been deployed to research institutions across Europe and North America, and the platform has been central to several EU-funded research projects including MEMMO and Horizon Europe initiatives.
PAL Robotics was founded in 2004 in Barcelona by a small group of engineers, including co-founder and CEO Francesco Ferro. The company traces its origins to a project building a robotic arm that could play chess for a client in the United Arab Emirates. In 2006, PAL Robotics launched REEM-A, the first full-size autonomous humanoid biped robot built in Europe, which was capable of bipedal walking, chess playing (using the Hydra chess engine), face recognition, and voice command processing.[1][2]
Over the following years, the company developed successive generations of humanoid platforms. REEM-B arrived in 2008 with enhanced locomotion. The REEM service robot debuted in 2012, designed for entertainment and public interaction. In 2013, REEM-C was introduced as a bipedal humanoid research platform with 44 degrees of freedom, improved mobility, and modular hardware suited for studying navigation, vision, artificial intelligence, and human-robot interaction.[3] REEM-C became widely used in academia but was position-controlled, which limited its capability for dynamic, force-sensitive tasks.
PAL Robotics has since expanded its product line to include TIAGo (a mobile manipulator launched in 2015), ARI (a social interaction robot introduced in 2019), and Kangaroo (a next-generation bipedal locomotion platform unveiled in 2021). The company is part of the PAL Technology Group based in Abu Dhabi, United Arab Emirates, and maintains offices in Barcelona, Toulouse, and Bari. As of late 2025, PAL Robotics employs approximately 148 people across five continents, with its robots serving customers in over 30 countries.[4]
TALOS was conceived to address the limitations of existing humanoid research platforms. The design was informed by collaboration between PAL Robotics and the Gepetto team at LAAS-CNRS (Laboratoire d'Analyse et d'Architecture des Systemes) in Toulouse, France. Led by researcher Olivier Stasse, the LAAS-CNRS team had extensive experience operating the Japanese HRP-2 humanoid and identified several shortcomings in that platform: insufficient motor power, overheating issues, and limitations in torque output from the motors and harmonic drives.[5]
The TALOS design aimed to resolve these problems by adopting a fully electric, torque-controlled architecture with high-bandwidth actuators and an industrial-grade EtherCAT communication bus. The first TALOS unit, named Pyrene, was delivered to LAAS-CNRS and officially presented in February 2017.[6] The robot was subsequently made commercially available later that year, with IEEE Spectrum reporting in March 2017 that PAL Robotics hoped TALOS would be "working side by side with humans" within five years.[7]
The founding paper, "TALOS: A new humanoid research platform targeted for industrial applications," was presented by Stasse et al. at the IEEE-RAS International Conference on Humanoid Robots (Humanoids) in 2017 and positioned TALOS alongside platforms such as Boston Dynamics' Atlas, NASA's Valkyrie, and Kawada's HRP-4 in the landscape of full-size humanoid research robots.[5]
TALOS is a fully electric biped humanoid robot built for robust, dynamic operation. The robot's 32 degrees of freedom are distributed across its body to provide human-like articulation:
| Body segment | Degrees of freedom |
|---|---|
| Each arm | 7 DOF |
| Each leg | 6 DOF (3-DOF hip, 1-DOF knee, 2-DOF ankle) |
| Waist/torso | 2 DOF |
| Neck/head | 2 DOF |
| Each hand/gripper | 1 DOF |
| Total | 32 DOF |
The 7-DOF arm configuration provides a large workspace with high payload limits. Each arm can lift 6 kg at full extension, a figure deliberately chosen to accommodate heavy industrial tools such as drills and riveters. The 6-DOF legs, combined with the 2-DOF waist, enable dynamic bipedal walking at speeds up to 3 km/h (approximately 1 m/s), navigation over irregular surfaces, and stair climbing.[8]
The robot's head is fully configurable, and the gripper end effectors can be customized to suit different research or application requirements. The overall design emphasizes modularity, allowing researchers to modify both hardware and software components.
| Parameter | Value |
|---|---|
| Height | 175 cm |
| Weight | 95 - 100 kg (configuration dependent) |
| Total degrees of freedom | 32 |
| Maximum walking speed | 3 km/h (1 m/s) |
| Arm payload (each arm, extended) | 6 kg |
| Stair climbing | Yes |
| Battery life | 1.5 hours (walking) / 3 hours (standby) |
| Onboard computing | 2x Intel Core i7 (COM Express Type-6) |
| Operating system | Ubuntu LTS with Real-Time OS |
| Software framework | ROS / ROS 2 |
| Control framework | ros_control with Whole Body Control |
| Communication bus | EtherCAT |
| Control loop frequency | 2 kHz standard; up to 5 kHz capable |
| Connectivity | WiFi, Ethernet, EtherCAT |
| Actuation | Fully electric (brushless DC motors) |
One of TALOS's defining features is its fully torque-controlled actuation system. The robot uses 32 high-torque brushless DC motors, each connected to a Harmonic Drive gear reducer, which is in turn connected to a torque sensor. Two high-precision encoders (19-bit resolution) are installed at each joint: one measuring the motor position and another measuring the joint position after the gear stage.[5] This dual-encoder arrangement, combined with joint-level torque sensing, provides the feedback necessary for compliant, force-sensitive control.
Torque sensors are integrated in all actuated joints except the head, wrists, and grippers. This comprehensive torque feedback enables multi-contact motions, compliant interaction with humans and the environment, and advanced whole-body control strategies that would be impossible with position-only feedback.[9]
The actuator design was motivated by experience with earlier platforms like HRP-2, where researchers encountered problems with power limitations, thermal overheating, and torque constraints from both the motors and the harmonic drives. TALOS's actuators were sized to provide sufficient torque margins for dynamic maneuvers including fast walking, stair climbing, and manipulation under load.[5]
TALOS is equipped with a comprehensive sensor package designed for robust perception and state estimation:
| Sensor type | Details |
|---|---|
| Joint torque sensors | In all actuated joints (except head, wrists, grippers) |
| Force/torque sensors | 6-axis sensors at both wrists and both ankles |
| Encoders | 19-bit high-precision encoders (motor and joint position) |
| IMU | Advanced Navigation Orientus AHRS (1 kHz measurement frequency) |
| Cameras | RGB-D cameras, stereo cameras |
| LiDAR | Yes |
| Temperature sensors | Monitoring at actuator level |
The integration of Advanced Navigation's Orientus sensor provides accurate tracking of the robot's base pose for balance and stability. The Orientus combines temperature-calibrated accelerometers, gyroscopes, and magnetometers with a fusion algorithm to deliver reliable orientation data at 1 kHz with low latency, which is critical for dynamic locomotion tasks such as walking on uneven terrain.[10]
State estimation fuses IMU data with forward kinematics to track the robot's floating-base pose in real time, providing the foundation for all higher-level control and planning algorithms.
TALOS uses an EtherCAT (Ethernet for Control Automation Technology) fieldbus for internal communication between its onboard computers and the distributed joint-level electronics. EtherCAT provides deterministic, low-latency data exchange, enabling the robot's control loops to run at 2 kHz in standard operation and up to 5 kHz when required. This high-frequency control is essential for torque-controlled walking, where rapid corrections are needed to maintain balance.[8]
The robot houses two Intel Core i7 processors (COM Express Type-6 modules) that handle real-time control, perception processing, and higher-level planning. The real-time control layer runs on an Ubuntu LTS installation with a real-time kernel, while perception and planning tasks can run on the standard Linux stack.
TALOS was designed from the start as a ROS-native platform. The robot's entire software stack is built on ROS, and it has been validated with multiple ROS distributions including Indigo, Kinetic, and Melodic. PAL Robotics also provides ROS 2 support, keeping the platform current with the latest developments in the ROS ecosystem.[11]
The low-level control system uses the ros_control framework developed by PAL Robotics, which provides a hardware abstraction layer for the joint-level torque, position, and velocity controllers. Through ros_control, researchers have direct access to encoder readings, IMU data, torque sensor measurements, force/torque sensor data, and temperature sensor readings from all instrumented joints.[8]
A key capability is the Whole Body Control (WBC) framework, which allows researchers to define tasks and constraints in a prioritized hierarchy. The WBC solver computes joint-level commands that satisfy the highest-priority constraints (such as maintaining balance) while optimizing for lower-priority objectives (such as tracking a desired hand trajectory).
PAL Robotics provides a complete Gazebo simulation environment for TALOS, available freely on the company's GitHub repositories. The simulation includes the robot's URDF model (with accurate center-of-mass, mass, and inertia tensor data), sensor plugins, and controller configurations. Researchers can launch the simulation with standard ROS commands and develop control algorithms in simulation before deploying them on the real hardware.[11]
The open-source Stack-of-Tasks (SoT) framework, maintained by the LAAS-CNRS Gepetto team, provides whole-body motion generation capabilities for TALOS. The talos-data and talos_robot packages on GitHub contain robot description files, configuration data, and integration code for connecting the SoT framework with TALOS through ros_control.[12]
Additional open-source packages include Pinocchio (a fast rigid-body dynamics library), Crocoddyl (an optimal control solver), and TSID (Task Space Inverse Dynamics), all of which have been validated on the TALOS platform and are widely used in the humanoid robotics research community.
TALOS has been a central platform in several European Union research projects:
MEMMO (Memory of Motion): The MEMMO project, funded by the EU's Horizon 2020 programme, aimed to develop a unified approach to motion generation for complex robots with arms and legs by building a "memory of motion" from pre-computed optimal trajectories. TALOS served as one of the primary experimental platforms for validating the project's algorithms. Research conducted under MEMMO included the development of the first whole-body Model Predictive Control (MPC) algorithm with full state feedback for a humanoid robot of TALOS's size and complexity. The MPC approach used "warm starts" from the motion memory to find optimal solutions within real-time constraints.[13]
PRIMI (Horizon Europe): The Kangaroo robot, TALOS's successor platform, participates in the PRIMI project (2023-2027), which aims to create robots that think, learn, and move with human-like precision and awareness. While PRIMI focuses on the Kangaroo hardware, the project builds directly on control and learning techniques pioneered on TALOS.[14]
PAL Robotics also participates in additional EU projects spanning smart cities, factories of the future, and artificial intelligence research, with TALOS and TIAGo frequently serving as experimental platforms.
As of the Humanoids 2020 workshop, at least six TALOS units were deployed worldwide at the following institutions:[9]
| Institution | Location | Primary research focus |
|---|---|---|
| LAAS-CNRS (Gepetto team) | Toulouse, France | Whole-body control, locomotion, motion planning |
| INRIA Nancy (LARSEN team) | Nancy, France | Teleoperation, machine learning, human-robot interaction |
| University of Edinburgh (SLMC group) | Edinburgh, UK | Multi-contact loco-manipulation, MPC, state estimation |
| University of Waterloo (RoboHub) | Waterloo, Canada | Gait synthesis, bimanual manipulation, HRI |
| Institut Jozef Stefan (IJS) | Ljubljana, Slovenia | Robotics research |
| PAL Robotics | Barcelona, Spain | Development, testing, demonstrations |
LAAS-CNRS: The Gepetto team at LAAS-CNRS, led by Olivier Stasse, received the first TALOS unit (Pyrene) in 2017 and has been the primary academic partner for the platform's development. Research at LAAS-CNRS has focused on whole-body inverse dynamics, optimal control for locomotion, and benchmarking different control architectures on the hardware.[5]
INRIA Nancy: The LARSEN project team at INRIA Nancy, led by Jean-Baptiste Mouret and Serena Ivaldi, uses TALOS for immersive teleoperation research. In this work, a human operator remotely controls the robot as an avatar that could be sent into dangerous situations. The whole-body teleoperation system tracks the operator's hand and head movements while a repulsor-based task prevents self-collisions and a force-based stabilizer corrects the robot's posture to prevent falls. The system adapts the operator's motions to account for the robot's own dynamics, strength, and weight, allowing the robot to imitate human movements without losing balance.[15]
University of Edinburgh: The Statistical Machine Learning and Motor Control (SLMC) group at the University of Edinburgh acquired a TALOS unit around 2020. Research at Edinburgh focuses on multi-contact loco-manipulation, hierarchical long-horizon motion planning, novel MPC techniques for hybrid problems (such as footstep selection and contact optimization), multi-objective trajectory optimization, and sensory integration strategies for state estimation and localization.[16]
University of Waterloo: The RoboHub facility at the University of Waterloo, housed in the Pearl Sullivan Engineering building, added TALOS to its research fleet. Programs supported by TALOS at Waterloo include control and gait synthesis, grasp and bimanual manipulation planning, human-robot interaction, and machine learning for robotics.[17]
TALOS has been the subject of numerous peer-reviewed publications. Notable research contributions include:
Benchmarking whole-body controllers: Ramuzat, Stasse, and Boria (2022) published a comprehensive comparison of three control schemes (position-based hierarchical QP, acceleration-level TSID, and torque-level TSID) applied to TALOS, evaluating tracking accuracy, energy consumption, and computational cost across flat terrain walking, uneven terrain walking, and stair climbing tasks.[18]
Whole-body teleoperation: Researchers at INRIA published preliminary results on whole-body teleoperation of TALOS, demonstrating safe remote control of the full humanoid body with fall prevention and self-collision avoidance.[15]
Actuator modeling and identification: Studies on actuator model identification and differential dynamic programming for TALOS have contributed to more accurate simulation-to-reality transfer for the platform.[19]
Reactive walking: Research at LAAS-CNRS on reactive walking for TALOS (Pyrene) has advanced the state of the art in real-time footstep adaptation and push recovery for humanoid robots.[20]
Model Predictive Control: The MEMMO project produced advances in whole-body MPC with full state feedback, using TALOS as the primary validation platform for algorithms that plan optimal motions in real time while respecting the robot's physical constraints.[13]
TALOS occupies a specific niche in the landscape of full-size humanoid research robots. Unlike Boston Dynamics' Atlas, which uses hydraulic actuation for high-bandwidth, high-power performance, TALOS is fully electric, resulting in quieter operation, lower friction, and simpler maintenance. However, Atlas can generate substantially higher forces and has demonstrated more dynamic maneuvers such as backflips and parkour.
Compared to NASA's Valkyrie (R5), TALOS offers a more accessible commercial platform with comprehensive ROS integration and open-source simulation tools. Valkyrie was designed primarily for the DARPA Robotics Challenge and Space Robotics Challenge, with a focus on disaster response and space exploration scenarios.
Relative to its predecessor REEM-C, TALOS represents a significant upgrade. While REEM-C has more degrees of freedom overall (44 DOF vs. 32 DOF), its position-controlled architecture limits its ability to perform force-sensitive tasks. TALOS's torque control, higher payload (6 kg per arm vs. 1 kg single-arm for REEM-C), and faster communication bus make it far more capable for dynamic locomotion and industrial manipulation research.
The Japanese HRP series (HRP-2, HRP-4, HRP-5P) from Kawada Industries and AIST represents the closest analog to TALOS in terms of research mission. TALOS was explicitly designed to address limitations discovered during years of operating HRP-2 at LAAS-CNRS, with improvements in actuator power, thermal management, and control bandwidth.[5]
| Robot | Manufacturer | Height (cm) | Weight (kg) | DOF | Actuation | Arm payload | Control frequency |
|---|---|---|---|---|---|---|---|
| TALOS | PAL Robotics | 175 | 95-100 | 32 | Electric (torque) | 6 kg/arm | Up to 5 kHz |
| Atlas | Boston Dynamics | 150 | 89 | 28 | Hydraulic | N/A | 1 kHz |
| Valkyrie (R5) | NASA JSC | 182 | 125 | 44 | Electric | N/A | N/A |
| REEM-C | PAL Robotics | 165 | 80 | 44 | Electric (position) | 1 kg/arm | N/A |
| HRP-4 | Kawada / AIST | 151 | 39 | 34 | Electric | N/A | N/A |
TALOS sits within a broader ecosystem of robots developed by PAL Robotics:
| Robot | Year | Type | Description |
|---|---|---|---|
| REEM-A | 2006 | Humanoid biped | First European full-size autonomous humanoid; chess-playing capability |
| REEM-B | 2008 | Humanoid biped | Enhanced locomotion and autonomy |
| REEM | 2012 | Wheeled humanoid | Service robot for entertainment and public interaction |
| REEM-C | 2013 | Humanoid biped | 44-DOF research platform for HRI, navigation, and AI |
| TIAGo | 2015 | Mobile manipulator | Versatile platform for logistics, healthcare, and research |
| TALOS | 2017 | Humanoid biped | Torque-controlled high-performance research platform |
| ARI | 2019 | Social robot | Interactive services with expressive face and conversation |
| Kangaroo | 2021 | Humanoid biped | Lightweight (40 kg) bipedal locomotion platform with jumping capability |
| TIAGo Pro | 2023 | Mobile manipulator | Omnidirectional mobility with force-sensing arms |
The Kangaroo platform represents PAL Robotics' next-generation approach to humanoid locomotion. At 160 cm tall and 40 kg, Kangaroo is significantly lighter than TALOS and uses custom force-controlled linear actuators (rather than rotary motors with harmonic drives) to achieve highly dynamic motions including running and jumping. Kangaroo is available in lite, standard, and pro variants and is designed as a headless platform optimized for industrial environments where social interaction is unnecessary.[14]
PAL Robotics forms part of the PAL Technology Group, headquartered in Abu Dhabi, United Arab Emirates. Under this structure, PAL Robotics has announced plans to open a robot manufacturing factory in Abu Dhabi, expanding its production capabilities beyond Barcelona.[21] The company continues to operate its primary R&D and engineering functions from its Barcelona headquarters.