PhoneBot: A Low-Cost Open Humanoid Robot Platform Reusing Smartphones

Ruochen Hou; Quanyou Wang; Daniel Koh; Dennis W. Hong

cs.ROOct 6, 2026

PhoneBot: A Low-Cost Open Humanoid Robot Platform Reusing Smartphones

Authors: Ruochen Hou, Quanyou Wang, Daniel Koh, Dennis W. Hong

Organizations: Robotics and Mechanisms Laboratory (RoMeLa), Department of Mechanical and Aerospace Engineering, University of California, Los Angeles, CA 90095, USA.

Abstract

The adoption of humanoid robots in education and research remains limited by high hardware costs, complex sensing systems, and substantial computational requirements. This paper presents PhoneBot, a low-cost, open-source humanoid robot platform that repurposes commodity smartphones as its primary sensing and computing unit. By using a smartphone's integrated inertial measurement unit (IMU), camera, wireless connectivity, and onboard processing capabilities, PhoneBot reduces hardware costs and simplifies the system architecture. The robot combines a modular lower-body structure driven by 13 low-cost actuators with a torso-mounted smartphone that supports perception, control computation, and user interaction. We describe the mechanical design, software architecture, and real-time communication framework that support stable locomotion and capabilities including vision-based human following, conversational interaction, filming, and mobile telepresence. Experimental evaluations demonstrate reliable walking, perception-driven interaction, and straightforward deployment using off-the-shelf consumer smartphones. With fully open-source hardware and software designs, PhoneBot provides an affordable, reproducible platform for education, research, and rapid prototyping. More details are available at https://phonebot.dev.

Figures & tables

Fig. 1 : PhoneBot is an open-source platform for bipedal robotics research that uses a commodity smartphone as its onboard computer to reduce hardware cost and simplify system architecture.

	ROBOTIS OP3 [ 3 ]	ToddlerBot [ 1 ]	BRUCE [ 4 ]	Berkeley Lite [ 2 ]	HighTorque MiniPi [ 5 ]	Ours
Height (m)	0.51	0.56	0.70	0.80	0.52	0.48
Weight	3.5	3.4	4.8	16.0	7.0	1.8
Active DoFs	20	30	16	22	12	13
Locomotion	✓	✓	✓	✓	✓	✓
Perception	✓	✓	✗	✗	✗	✓
Open Source	Design & Code	Design & Code	Code Only	Design & Code	Code Only	Design & Code

TABLE I : Comparison with Small-Scale Humanoid Research Platforms.

Fig. 2 : Joint configuration and CoM of PhoneBot. (a) Front view. (b) Left view.

Name	Joint	Actuator	Range ( ∘ )
Hip yaw	J1, J7	XL430-W250	$[-90,\,90]$
Hip roll	J2, J8	XL430-W250	$[-90,\,90]$
Hip pitch	J3	XL430-W250	$[-90,\,123]$
Hip pitch	J9	XL430-W250	$[-123,\,90]$
Knee pitch	J4, J10	XL430-W250	$[-90,\,90]$
Ankle pitch	J5, J11	XL430-W250	$[-90,\,90]$

TABLE II : Joint specifications and Dynamixel motor assignments.

Gear Ratio	Stall Torque (N m)	No-Load Speed (RPM)	Stall Current (A)
$258.5{:}1$	$1.4$	$57$	$1.3$
Voltage (V)	Resolution (Pulse/Rev)	Mass (g)	Dimensions (mm)
$11.1$	$4096$	$57.2$	$46.5\times 34.0\times 28.5$

TABLE III : Manufacturer specifications of the DYNAMIXEL XL430-W250 actuator.

Fig. 3 : System overview of PhoneBot

Parameter	Symbol	Value
Firmware / sim $k_{p}$ scale	$r_{p}$	$150$
Firmware / sim $k_{d}$ scale	$r_{d}$	$16$
Firmware position gain	$k_{p}^{\mathrm{real}}$	$600$
Firmware velocity gain	$k_{d}^{\mathrm{real}}$	$0$
Simulator stiffness	$k_{p}$	$4.0~\mathrm{N{\cdot}m/rad}$
Simulator damping	$k_{d}$	$0$

TABLE IV : Dynamixel XL430 actuation parameters used in the torque-aware simulator. All 13 joints share this model.

Component	Dim.	Actor	Critic
Angular velocity (noisy)	$3$	$\checkmark$	$\checkmark$
Gravity vector (noisy)	$3$	$\checkmark$	$\checkmark$
Velocity command	$3$	$\checkmark$	$\checkmark$
Joint position (noisy, rel. home)	$13$	$\checkmark$	$\checkmark$
Joint velocity (noisy)	$13$	$\checkmark$	$\checkmark$
Previous action	$13$	$\checkmark$	$\checkmark$

TABLE V : Observation space of the torque-aware walk policy. Actor inputs are the onboard signals used at deployment. Critic inputs add privileged simulation state.

Term	Equation	Weight
Linear velocity tracking	$\exp\bigl(-\\|v_{xy}-c_{xy}\\|_{2}^{2}/\sigma\bigr)$	$4.0$
Yaw velocity tracking	$\exp\bigl(-(\omega_{z}-c_{z})^{2}/\sigma\bigr)$	$4.0$
Alive	$1$	$0.5$
Vertical velocity	$v_{z}^{2}$	$-0.2$
Roll/pitch rate	$\\|\omega_{xy}\\|_{2}^{2}$	$-0.1$
Orientation	$\\|g_{xy}\\|_{2}^{2}$	$-10.0$

TABLE VI : Reward terms and weights of the torque-aware walk policy. Velocities are IMU-local.

\sigma=0.1

h^{*}=0.03~\mathrm{m}

u=|\tau|/\tau_{\mathrm{lim}}

, and

\mathbb{I}_{\mathrm{move}}=\mathbf{1}_{\|c\|>0.05}

Term	Equation	Weight
Height	$\exp(-(z-z^{\star})^{2}/\sigma_{z}^{2})$	$1.0$
Height (sharp)	$\exp(-(z-z^{\star})^{2}/\sigma_{z,\mathrm{s}}^{2})$	$1.0$
Height $\ell_{1}$	$-\|z-z^{\star}\|$	$7.5$
Upward velocity	$\max(v_{z},0)\,\mathrm{clip}((z^{\star}-z)/0.02,0,1)$	$0.75$
Upright linear	$g_{z}$	$1.5$
Upright sharp	$\exp(-(\\|g_{xy}\\|/\sigma_{u})^{2})$ gated by height	$1.5$

TABLE VII : Standup reward terms and weights. The target height is

z^{\star}=0.282~\mathrm{m}

. The terms

h_{c}

u_{c}

, and

p_{c}

are Gaussian height, uprightness, and pose scores; each

\sigma

is the width of its corresponding Gaussian.

Term	Equation	Weight
Look yaw	$\exp(-(\theta/\sigma_{\theta})^{2})$	$1.5$
Look elevation	$\exp(-(\varphi/\sigma_{\varphi})^{2})$	$1.0$
Look product	$r_{\mathrm{yaw}}\,r_{\mathrm{elev}}$	$1.0$
Pelvis look yaw	$\exp(-(\psi/\sigma_{\theta})^{2})$	$2.0$
Zero IMU $v_{xy}$	$\exp(-\\|v_{xy}\\|_{2}^{2}/\sigma)\,(1-\mathbb{I}_{\mathrm{step}})$	$4.0$
Stepping tax	$\mathbb{I}_{\mathrm{step}}$	$-0.5$

TABLE VIII : Heading reward terms and weights. The terms

\theta

and

\varphi

are remaining look yaw and elevation in the IMU frame, and

\psi

is look yaw in the pelvis frame. Gait terms while stepping match Table VI .

Fig. 4 : Evaluation curves for the torque-aware walk policy trained from scratch with sagittal mirroring (blue) and without it (red). Both methods use the same reward weights from Table VI . Each solid line is the mean over five seeds (

0

–

4

). The shaded region is

\pm 1

standard deviation across seeds. Panel (a) is the episode return: the weighted sum of the other terms, multiplied by the control timestep

\Delta t=0.02~\mathrm{s}

and accumulated over the episode. Panels (b)–(l) show the corresponding weighted terms before that

\Delta t

factor. The horizontal axis is environment steps, shown up to

1.5\times 10^{8}

Fig. 5 : Deterministic gait symmetry at approximately

4.5\times 10^{8}

environment steps. (a) Time-averaged joint-angle RMSE between one rollout and the sagittally mirrored phase-swapped rollout. (b) Relative left–right difference in mean absolute joint torque. (c) Relative left–right difference in mean absolute mechanical power

|\tau\dot{q}|

. Points are seeds; black markers show mean

\pm 1

standard deviation.

Fig. 6 : Physical demonstration of PhoneBot capabilities. (a)Bipedal locomotion. (b) Human tracking with front camera. (c) Stand up policy from face-down to fully upright stance.

	Hardware					CPU usage
Phone	Price	Released	CPU	RAM	Camera	Idle	IMU $200$ Hz	Policy	Vision detection	SLAM	HRI
Honor 9	used (2017)	Jul. 2017	Kirin 960	4/6 GB	$12{+}20$ MP	$1.1\%$	$12.1\%$	$9.8\%$	$17.9\%$	$\times$	$2.5\%$
moto g 2025	$50	Jan. 2025	Dimensity 6300	4 GB	$50{+}2$ MP	$0.4\%$	$13.4\%$	$10.3\%$	$15.6\%$	$34.3\%$	$3.6\%$
Galaxy A16 5G	$50–$ 80	Oct. 2024	Exynos 1330	4 GB	$50{+}5{+}2$ MP	$0.8\%$	$8.5\%$	$7.1\%$	$17.6\%$	$28.3\%$	$5.2\%$
moto g 2024	$50–$ 90	Mar. 2024	Snapdragon 4 Gen 1	4 GB	$50{+}2$ MP	$0.8\%$	$9.1\%$	$6.3\%$	$16.7\%$	$25.2\%$	$4.3\%$

TABLE IX : Hardware and onboard CPU usage of the four test phones. GPU is omitted because the current application stack is CPU-bound. Idle is the app with no function running. SLAM on Honor 9 is marked

\times

because Google services are unavailable on that Huawei device. Prepaid prices are carrier-locked listings (Sep. 2026) except Honor 9, which is a used 2017 unit (China launch about $340).

Explore similar work

May 15, 2026cs.RO

Nori Bot: A Sub-$1,000 Floor-to-Counter Mobile Manipulator

Open-source mobile manipulators have reached

660 (XLeRobot) but every sub-

1,000 platform shares three limitations: a fixed-height workspace, reactive-only control, and no protection against the stall-induced burn-out that destroys cheap Feetech servos. We present Nori Bot, a 17-DoF dual-arm mobile manipulator at $947 (~3% the cost of comparable commercial platforms) that addresses all three: (1) a 600mm Z-axis lift on the existing servo bus for floor-to-counter reach; (2) a thin-client Raspberry Pi 4 paired with the OpenClaw proactive agent runtime so cron jobs and hooks trigger physical tasks autonomously; and (3) a software safety stack with sensorless grip-force feedback via motor current on a soft TPU finger. Code, CAD, and the skill manifest will be released.

Antonio Li, Sungjoon Park, Wen Ni Chew

Department of Computer Science, Columbia University

Robocasa Floor Openclaw

Ask PDF

May 3, 2026cs.RO

Phone2Act: A Low-Cost, Hardware-Agnostic Teleoperation System for Scalable VLA Data Collection

Collecting diverse, high-quality manipulation data for Vision-Language-Action (VLA) model training remains prohibitively expensive for many research groups, as existing teleoperation frameworks rely on specialized hardware or are tightly coupled to specific robot platforms. We present Phone2Act, a low-cost, hardware-agnostic teleoperation framework that transforms a commodity smartphone into a 6-DoF robot controller via Google ARCore. Built on a modular ROS 2 architecture, Phone2Act decouples control logic from hardware specifics through interchangeable bridge nodes, supporting platforms from industrial cobots to low-cost bimanual arms without code modification. A Universal Recorder synchronizes multi-camera RGB streams with robot state feedback and exports demonstrations natively in the LeRobot dataset format, eliminating post-processing and enabling immediate VLA fine-tuning. We validate the framework by fine-tuning GR00T-N1.5 on 130 collected episodes, achieving a 90% success rate on a real-world multi-stage pick-and-place task deployed on a physical Dobot CR5.

Om Mandhane, Bipin Yadav, Sangeetha Prasanna Ram +1

Dept. of Automation & Robotics Engineering, Vivekanand Education Society’s Institute of Technology (VESIT), Mumbai, India.

Diffusion-Based Vision-Language-Actions Teleoperation Robotic Data Acquisition Dial

Ask PDF

Sep 16, 2026cs.RO

Hardware-Free Robotics Laboratories in Mixed Reality

Teaching robotics relies on screen-based simulation, showing robot motion in an abstract coordinate frame rather than at real scale in the learner's own space, while access to physical hardware is limited by cost, safety, and scheduling constraints. We present MR-Robotics LAB, a mixed-reality (MR) platform that replays MATLAB-generated robot trajectories at real scale within the learner's physical environment. A browser-based service validates a MATLAB workspace file (.mat), normalizes units, and publishes a versioned JSON trajectory; a Unity application on a Meta Quest 3 then reproduces the authored joint configurations under position control and replays them at the declared frame rate within a physics-enabled scene that supports collision detection and end-effector grasping. A formative single-group evaluation with engineering students found that participants reported low setup effort (M = 4.67 on a 5-point scale) and perceived support for workspace understanding from multi-viewpoint inspection (M = 4.56), and 83% of participants affirmed their willingness to use the platform in an introductory robotics course. The evaluation instrument records only perceived outcomes, without counterbalancing or a learning measure, so no comparative advantage over desktop simulation is claimed. The contribution is a reusable simulation-to-MR trajectory pathway and design guidance for hardware-free robot visualization in engineering education.

Santiago Berrezueta-Guzman, Habiba-Loai Khalil, Andrei Koshelev +2

Technical University of Munich, Heilbronn, Germany

Robot Autonomy Robot Trajectories Laboratory

Ask PDF