Franka Emika Panda

Home > Hardware > Arms > Franka Panda

Overview

Franka Emika Panda is a 7-axis collaborative robot (cobot) developed in Munich, Germany. With torque sensors embedded in all joints, it implements human-level tactile sensing and force control, making it an innovative research robot platform. Selected as one of Time magazine’s “50 Best Inventions” in 2018, it has become the de facto standard platform for VLA (Vision-Language-Action) and robot learning research.

Core Specifications Table

Item	Franka Panda	Franka Research 3 (FR3)
Manufacturer	Franka Emika (Germany, now Agile Robots)	Franka Emika
Degrees of Freedom (DoF)	7-axis	7-axis
Payload	3 kg	3 kg
Reach	850 mm	855 mm
Repeatability	±0.1 mm	±0.1 mm
Maximum Speed	2 m/s (end effector)	2 m/s
Robot Weight	18 kg	18 kg
Control Frequency	1 kHz	1 kHz
Force Resolution	< 0.05 N	< 0.05 N
Force Accuracy	0.8 N (relative)	0.8 N
Force Repeatability	0.15 N	0.15 N
Workspace Coverage	-	94.5%
Price	~$30K+ (educational discounts available)	~$40K+

Key Significance

1. Democratization of Research Robots

At launch, Franka Panda disrupted the market dominated by large, heavy, and expensive industrial robots with its revolutionary price of approximately $10,500. This enabled precision manipulation research at universities and research institutes worldwide, with world-leading institutions such as MIT, Stanford, and Oxford Robotics Institute adopting it as their standard platform.

2. Data Collection Standard for VLA/Robot Learning

Used as a core collection platform for large-scale robot learning datasets including Open X-Embodiment, DROID, and BridgeData, it has become the foundation for training state-of-the-art VLA models like OpenVLA, RT-X, and Octo.

3. Popularization of Torque Sensor-Based Force Control

While traditional industrial robots focused on position control, Franka Panda embeds torque sensors in all 7 joints, enabling real-time impedance control and collision detection.

Design Philosophy

”Robotics for Everyone”

Franka Emika founders Sami Haddadin and Simon Haddadin designed Panda with the vision of “robots that safely collaborate with humans”.

Core Design Principles

Human-Centric Design
- 7-axis design implements kinematic structure similar to human arm
- Lightweight design at 18kg, 3.6 times human arm weight (approximately 5kg)
- Designed as collaborative robot, capable of close proximity work with humans when proper risk assessment and safety settings are applied (ISO 10218, ISO/TS 15066 compliance required)
Implementation of Tactile Sensing
- Torque sensors embedded in all joints (Link-side torque sensing)
- Force resolution below 0.05N for detecting fine contacts
- Real-time external force estimation and collision detection algorithms
Research-Friendly Openness
- Franka Control Interface (FCI): 1kHz real-time control API
- Real-time access to joint position, velocity, torque, external force estimates
- Provides inertia matrix, gravity vector, Jacobian, Coriolis terms
- ROS, ROS2, MATLAB/Simulink integration support
Torque Control First Philosophy
- Traditional robots: Current overload check → Difficult to sense force
- Franka: Strain gauge-based direct torque measurement → Precise force control
- Compliance control enables flexible object manipulation

DLR Legacy

Franka’s technical roots lie in the German Aerospace Center (DLR) lightweight robot (LWR) research. Sami Haddadin researched human-robot interaction safety at DLR, and this experience is directly reflected in Panda’s collision detection and response algorithms.

Technical Details

Torque Sensor System

Joint 1-7: Link-side Torque Sensors

End-effector Cartesian force estimation performance (per official datasheet):

Force Resolution: < 0.05 N
Relative Force Accuracy: 0.8 N
Force Repeatability: 0.15 N

Real-time Data (1 kHz):

Joint positions (q)
Joint velocities (dq)
Link-side torque (tau_ext)
External force estimation (F_ext)
Inertia matrix M(q)
Coriolis term C(q, dq)
Gravity vector g(q)
Jacobian J(q)

Control Interface (FCI)

Franka Control Interface provides the following through libfranka library:

Position Control: Joint space and Cartesian space
Velocity Control: Direct joint velocity command
Torque Control: Direct joint torque command (impedance control implementation)
Hybrid Control: Position/force hybrid control

VLA Research Applications

Franka Panda is one of the most widely used robot platforms in VLA (Vision-Language-Action) model research.

Major Datasets

Dataset	Scale	Franka Usage	Description
Open X-Embodiment	1M+ episodes	Major source	22 robot types, Franka is core
DROID	76,000+ demos	Core platform	Various manipulation demonstrations
BridgeData V2	60,000+ trajectories	Partially included	Tabletop manipulation data
LIBERO	130 tasks	Simulation	PyBullet-based Franka simulator
Franka Kitchen	Various tasks	Dedicated	dm_control-based kitchen environment
panda-gym	5 tasks	Dedicated	OpenAI Gym integrated RL environment

Major VLA Models and Papers

OpenVLA (2024)

Trained on Open X-Embodiment 970K trajectories
Demonstrates fine-tuned policy on Franka Panda video
Standard benchmark for open-source VLA models

ReBot (2025)

Real-to-Sim-to-Real data augmentation technique
17% improvement on Octo, 20% improvement on OpenVLA in Franka Panda experiments

CoT-VLA (2025)

Introduces Visual Chain-of-Thought reasoning
46.7% relative performance improvement on Franka-Tabletop demo fine-tuning (53.7% → 78.8%)

LoLA (2025)

Long Horizon Latent Action Learning
LIBERO benchmark and real Franka robot evaluation
Introduces State-Aware Latent Re-representation module

RT-X / Octo

Google DeepMind’s Robot Transformer series
Open X-Embodiment based, utilizes Franka data

Simulation Environments

LIBERO: PyBullet-based long-horizon language-conditioned manipulation benchmark
panda-gym: OpenAI Gym compatible RL environment (reach, push, slide, pick & place, stack)
RLBench: CoppeliaSim-based 100+ task benchmark
Isaac Sim: NVIDIA Omniverse-based Franka USD assets provided

Comparison with Other Robots

Item	Franka Panda	UR5e	Sawyer	KUKA LWR
DoF	7	6	7	7
Payload	3 kg	5 kg	4 kg	7 kg
Reach	850 mm	850 mm	1,260 mm	800 mm
Repeatability	±0.1 mm	±0.03 mm	±0.1 mm	±0.1 mm
Torque Sensors	All joints	None	Per joint	All joints
Force Control	Native	External F/T required	Limited	Native
1kHz Control	Supported	Limited	Not supported	Supported
ROS Support	Excellent	Good	Discontinued	Good
Research Ecosystem	Largest	Moderate	Limited	Moderate
Price	~$30K+	~$35K+	Discontinued	~$100K+

Franka’s Differentiators

Native Torque Control: UR5 only performs current-based overload checking, making precise force tasks difficult
Research Ecosystem: De facto standard platform for state-of-the-art research including VLA and imitation learning
Open API: 1kHz real-time control and full dynamics model access
Safety: Advanced safety algorithms based on DLR collision research

Company History and Current Status

Founding Team

Sami Haddadin: Co-founder, Robotics/AI Professor (TUM MIRMI founding director, 2018-2025)
Simon Haddadin: Co-founder, CEO (physician background, DLR researcher)
Sven Parusel: Co-founder

Major Milestones

Year	Event
2016	Franka Emika GmbH founded (Munich)
2017	German Future Prize (Deutscher Zukunftspreis) winner
2018	Time “50 Best Inventions” selection, 2,000 units shipped in first year
2020	National Geographic feature
2022	Franka Research 3 (FR3) release
2023.08	Preliminary insolvency filing due to shareholder dispute
2023.11	Acquired by Agile Robots AG (approximately 32 million euros)

Current Status (2024-2025)

Ownership: Agile Robots AG (Munich, Softbank/Foxconn investment)
Employees: Approximately 100 maintained
Product Line: Franka Research 3, Franka Production 3 continue sales
Note: Legal disputes regarding patents occurred during acquisition process

Purchase and Implementation Guide

Pricing Information

Franka Research 3: $40,000+ (educational institution discounts available)
Franka Production 3: Industrial use, separate quote
Additional Costs: Gripper, mounting, training, support

Required Infrastructure

Stable mounting surface (minimize vibration)
Real-time Linux PC (Ubuntu recommended)
Network connection (Franka Control Interface)

Alternative Considerations

Research: Franka FR3 (when torque control needed), UR5e (versatility priority)
Educational: Start with panda-gym, LIBERO simulation
Low Budget: Used Panda, or simulation only