Autonomous Navigation Drone Setup and DIMOS Integration Testing

[dimos-bot]

1 Project Overview

This project aims to build a 10-inch drone platform with autonomous mapping, localization, path planning, and navigation capabilities, designed to validate the autonomous control abilities of Dimos-Agent over unmanned aerial vehicles.

The drone must be capable of autonomous indoor flight missions and support integration with ROS 2, PX4, and navigation algorithms (SLAM, Navigation), in order to verify Agent capabilities in task planning, flight control, and anomaly handling.

As this platform is primarily intended for algorithm validation and functional testing rather than a final product, it must meet the following requirements:

• Support autonomous mapping (SLAM) and real-time localization;
• Support autonomous navigation, waypoint flight, and obstacle avoidance;
• Stable flight suitable for low-speed indoor operation;
• Sufficient flight endurance to minimize test interruptions from frequent battery swaps;
• Compatible with the PX4 and ROS 2 ecosystem for ease of secondary development;
• Mature hardware with low maintenance costs and easily replaceable parts.

Since the Agent may execute insufficiently validated control strategies during development and debugging — posing risks of collision or loss of control — this platform is designated as a functional validation testbed. Hardware selection follows the principles of cost-effectiveness, easy procurement, simple maintenance, and abundant spare parts, minimizing overall cost while meeting navigation and flight requirements.

2 Hardware Selection Principles

The platform follows these selection criteria:

1. Maturity and Stability
   • Prioritize community-proven solutions compatible with PX4 and ROS 2.
   • Avoid experimental hardware wherever possible.
2. Ease of Repair
   • Motors, propellers, frames, and other wear parts can be quickly procured and replaced.
   • Use standard-sized components wherever possible.
3. Cost Control
   • Prioritize cost-effective products that meet autonomous navigation requirements.
   • Individual component failures should allow rapid replacement without disrupting development progress.
4. Expandability
   • Reserve computational headroom, power budget, and mounting space for future additions such as LiDAR, stereo cameras, or robotic arms.
5. Indoor Navigation Friendly
   • The flight platform should have good hover stability.
   • Sufficient payload capacity to carry the compute platform and additional sensors.

3 Hardware Selection

Component	Recommended Model	Rationale
Flight Controller	MicoAir Tech MicroAir743 V2	Excellent PX4 compatibility, large user base, mature ecosystem; manufacturer based in Shenzhen for timely technical support.
ESC	FlyingRC 4-in-1 PDB + DJI Phantom ESC	PDB integrates 5V/5A and 12V/4A BEC for direct onboard power; DJI Phantom ESCs include filtering and soft-braking to suppress voltage spikes, more stable than typical FPV ESCs.
Receiver	MicoAir Tech TRS 2-in-1 Receiver	Integrates RC reception and video transmission, reducing module count and wiring complexity.
RF Module	MicoAir Tech TRS RF Head	Paired with the TRS receiver to ensure stable RC and video link.
Remote Controller	RadioMaster TX15 / Pocket	Supports RF module installation, seamlessly compatible with the MicoAir Tech TRS system; large community, flexible configuration.
Frame	Custom 10-inch 3D-Printed Frame (440 mm wheelbase)	All structural parts are 3D-printable for flexible design; damaged parts can be reprinted quickly at low cost, enabling fast iteration.
Motors	2212 / 2216 Brushless Motors (any brand)	High market availability, easy to source, low cost — ideal for a functional validation platform.
Propellers	EOLO 1038 Self-locking Props	10-inch props offer significantly higher thrust efficiency than 5-inch; good aerodynamic balance, self-locking design for easy installation; low cost and widely available.
Battery	4S2P 21700 (EVE JP40 / EEMB 50PL)	Energy density > 220 Wh/kg, ~20C continuous discharge; balances thrust and endurance. Can be upgraded to Changyun semi-solid-state (~330 Wh/kg) for maximum flight time.
LiDAR	Livox Mid-360	Well-documented, mature ROS/PX4 ecosystem, excellent mapping and localization performance for indoor autonomous navigation.
Depth Camera (alt.)	Intel RealSense D435	Well-documented, mature ROS/PX4 ecosystem; suitable for low-cost visual localization validation.
Onboard Computer	Jetson Orin NX 16GB + Damiao Carrier Board	Ample compute for ROS 2, SLAM, navigation, and Agent inference; large user base, mature ecosystem, Shenzhen-based vendor for convenient support.
Camera	USB Webcam	Meets image capture and vision validation needs; low cost, easy deployment, upgradeable later.
GPS (optional)	MicoAir Tech M9	Stable performance, easy to source.
4G Module (optional)	Hangzhou Yexin CM4 MINI 4G Module	Qualcomm chipset, reliable connectivity.
Gimbal (optional)	XFly Z2 Mini Dual-Sensor Gimbal	Existing equipment; supports visible/thermal imaging validation — not required for navigation but installable for specific missions.

Two configurations are available depending on the current development phase: the full configuration (Jetson Orin NX 16GB + Livox Mid-360) for high-precision LiDAR navigation validation, and the lightweight configuration (Jetson Orin NX 8GB + Intel RealSense D435) as a low-cost indoor/outdoor full-feature Agent validation platform. Both use mature, readily available, and easily maintained flight hardware.

4 Technology Stack

4.1 Flight Controller

Selection: PX4

PX4 is one of the most mature open-source autopilot systems available, offering comprehensive flight control, broad hardware support, and an active development community. It supports ROS 2, MAVLink, and Micro XRCE-DDS for communication, enabling seamless integration with autonomous navigation and Dimos-Agent systems. PX4 also provides mature simulation environments, log analysis tools, and strong extensibility to support long-term project iteration.

4.2 Robot Middleware

Selection: ROS 2 Humble

ROS 2 has become the mainstream software framework for robotics, supporting distributed communication, lifecycle management, and component-based development. It is highly compatible with PX4, autonomous navigation algorithms, and visual perception systems. The project adopts ROS 2 Humble as the unified middleware layer, facilitating integration of SLAM, navigation, visual perception, and Dimos-Agent modules.

4.3 Autonomous Navigation Framework

Selection: EGO-Planner-Swarm (ROS 2 version)

EGO-Planner has been validated through years of engineering practice and has a mature ROS 2 version (ego-planner-swarm) with an active community, comprehensive documentation, and abundant deployment experience. Diff-Planner, also from ZJU FAST Lab, introduces diffusion models into trajectory planning for improved performance in complex environments, but its public engineering work is primarily ROS 1-based and engineering maturity is still evolving. SUPER, from HKU MaRS Lab, represents a next-generation navigation framework with advances in mapping, environment representation, and trajectory optimization, but it currently has higher integration complexity and fewer ROS 2 deployment examples.

Considering ROS 2 support, community maturity, engineering experience, development risk, and maintenance cost, the project adopts EGO-Planner-Swarm (ROS 2 version) in Phase 1 to validate Dimos-Agent's autonomous mapping, localization, navigation, and control capabilities. Future migration to SUPER or Diff-Planner can be evaluated once the platform is stable.

4.4 Agent Framework

Selection: DimOS

This project adopts DimOS as the intelligent decision-making and task execution framework. DimOS is an Agent Operating System designed for robotic systems, supporting natural language task planning, skill invocation, multi-agent collaboration, and spatial memory. It serves as the autonomous decision layer for the drone, decoupled from the navigation, perception, and flight control modules.

While DimOS officially supports operation without ROS, it also provides complete ROS 2 integration, allowing direct reuse of existing ROS 2 navigation, SLAM, and perception algorithms. The project therefore adopts a PX4 + ROS 2 + EGO-Planner + DimOS software architecture: PX4 handles low-level flight control, ROS 2 handles perception/localization/navigation communication, and DimOS handles task understanding, decision planning, and skill scheduling — calling navigation and flight control capabilities via ROS 2 interfaces to achieve autonomous Agent control of the drone.

4.5 Software Architecture

The project adopts a layered software architecture, from top to bottom: Task Decision Layer (DimOS), Navigation & Perception Layer (ROS 2 Humble), Flight Controller Communication Layer (MAVROS), Flight Controller Layer (PX4), and Actuator Layer (motors/ESCs). Each layer is decoupled through standard interfaces for independent iteration and replacement.

┌──────────────────────────────────────────────────────────────────┐
│                      Task Decision Layer                         │
│                    DimOS Agent Framework                         │
│   NL Understanding | Task Planning | Skill Dispatch | Exception  │
└─────────────────────────────┬────────────────────────────────────┘
                              │ ROS 2 Interface (Topic / Service / Action)
┌─────────────────────────────▼────────────────────────────────────┐
│              Navigation & Perception Layer (ROS 2 Humble)        │
│                                                                  │
│  ┌──────────────────┐  ┌─────────────────┐  ┌────────────────┐  │
│  │ EGO-Planner-     │  │   FAST-LIO2     │  │ OctoMap        │  │
│  │ Swarm (ROS2)     │◄─│  Localization   │  │ Mapping        │  │
│  │ Planning/Avoid.  │  │  + SLAM         │  │ Obstacle Map   │  │
│  └────────┬─────────┘  └────────▲────────┘  └───────▲────────┘  │
│           │                     │PointCloud2 / IMU   │           │
│           │            ┌────────┴────────────────────┘           │
│           │            │  livox_ros_driver2                      │
│           │            │  (Mid-360 Driver, ROS 2)                │
│           │            └─────────────────────────────────────────┤
└───────────┼──────────────────────────────────────────────────────┘
            │ Control Commands (position / velocity / attitude)
┌───────────▼──────────────────────────────────────────────────────┐
│                 FC Communication Layer                            │
│                  MAVROS (ROS 2 Humble)                           │
│                     MAVLink Protocol                             │
└───────────┬──────────────────────────────────────────────────────┘
            │ UART (Orin NX → Pixhawk TELEM2, 921600 baud)
┌───────────▼──────────────────────────────────────────────────────┐
│                    Flight Controller Layer                        │
│                      PX4 Autopilot                               │
│   Attitude Control | Motor Mixing | Failsafe | Sensor Fusion     │
└───────────┬──────────────────────────────────────────────────────┘
            │ PWM / DSHOT
┌───────────▼──────────────────────────────────────────────────────┐
│                       Actuator Layer                             │
│              ESC + Brushless Motors + Propellers (10-inch)       │
└──────────────────────────────────────────────────────────────────┘
Hardware Connections:
  Jetson Orin NX ←── Ethernet ──► Livox Mid-360 (point cloud + IMU)
  Jetson Orin NX ←── UART/USB ──► PX4 Flight Controller (MAVLink)

5 Performance Estimates

5.1 Flight Endurance Estimate

Battery Parameters

The platform uses EVE JP40 21700 cells in a 4S2P configuration (8 cells total). Measured parameters:

Parameter	Value
Cell Model	EVE JP40 21700
Configuration	4S2P (8 cells)
Measured Capacity per Cell (2C CC discharge)	3900 mAh
Total Capacity (2P)	7800 mAh
Average Discharge Voltage	3.66 V/cellCalculation* Required hover thrust: 1178 g * Single motor hover load: 1178 ÷ 4 ≈ 295
Pack Nominal Voltage	14.64 V (4S × 3.66 V)
Energy per Cell (measured)	14.274 Wh
Total Pack Energy	114.2 Wh (14.274 × 8)
Battery Weight	570 g
Cell Energy Density	204 Wh/kg

Total Weight Breakdown

Component	Weight
Airframe (frame, motors, ESCs, FC, etc.)	1200 g
Livox Mid-360 LiDAR	300 g
4S2P Battery	570 g
Total	2070 g

Hover Endurance Calculation

• Required hover thrust: 2070 g (equal to total weight)
• Single motor hover load: 2070 ÷ 4 ≈ 518 g
• Motor thrust efficiency (measured): 6.89 g/W
• Hover power (motors): 2070 ÷ 6.89 ≈ 300.4 W
• Onboard electronics (Jetson Orin NX + LiDAR + misc.): 50 W
• Combined total power: 300.4 + 50 = 350.4 W
• Theoretical hover time: 114.2 Wh ÷ 350.4 W × 60 ≈ 19.6 min
• Corrected theoretical endurance (× 0.9): ≈ 17.6 min

Accounting for navigation maneuvers, acceleration/deceleration, wiring losses, and ESC efficiency, estimated practical flight endurance is 15–17 min.

5.2 Localization Performance (FAST-LIO2 + Livox Mid-360)

FAST-LIO2 is a leading LiDAR-Inertial odometry solution for indoor environments. The Livox Mid-360 provides a 360° horizontal and 59° vertical field of view. Together they achieve excellent localization and mapping accuracy in structured indoor scenes.

Metric	Estimated Value	Notes
Localization Accuracy (RMSE)	< 3 cm	Structured indoor environment, 10 m range
Mapping Accuracy	< 5 cm	Loop closure error within 20 m range
Odometry Output Rate	100 Hz	Synchronized with IMU frequency
Point Cloud Map Rate	10 Hz	Limited by Mid-360 scan frame rate
Localization Compute Latency	< 10 ms	Measured on Orin NX
Applicable Scenes	Indoor, corridors, semi-open spaces	Accuracy degrades in feature-sparse scenes

Note: In feature-sparse environments such as plain white rooms or large glass/mirror surfaces, pure LiDAR localization accuracy may degrade. IMU pre-integration is recommended to maintain short-term stability during feature-loss periods.

5.3 Navigation and Planning Performance (EGO-Planner-Swarm)

EGO-Planner-Swarm uses ESDF-free gradient-based trajectory optimization with very low computational overhead, suitable for real-time execution on the Jetson Orin NX. The following data is sourced from ZJU FAST Lab publications and community real-world flight results:

Metric	Reference Value	Notes
Single Planning Time	< 2 ms	Single CPU core; far below ESDF-based ~50 ms
Planning Loop Rate	50 Hz	Real-time response to environmental changes
Max Flight Speed (obstacle avoidance mode)	3–5 m/s	Measured in complex indoor obstacle environments
Trajectory Tracking Position Error	< 0.2 m	Speed ≤ 3 m/s, indoor scenes
Minimum Safe Obstacle Clearance	0.5 m	Depends on inflation_radius parameter
Map Update Rate	10 Hz	Limited by LiDAR frame rate
Re-planning Response Time	< 20 ms	Triggered upon detection of new obstacles

5.4 Compute Resource Usage Estimate (Jetson Orin NX 16GB)

The Jetson Orin NX 16GB provides up to 100 TOPS AI compute and an 8-core ARM CPU. Estimated full-stack runtime resource usage:

Module	CPU Usage (est.)	Memory Usage (est.)
livox_ros_driver2	< 5%	< 200 MB
FAST-LIO2	15–25%	< 500 MB
EGO-Planner-Swarm	10–15%	< 300 MB
MAVROS	< 5%	< 100 MB
ROS 2 Communication Infrastructure	< 5%	< 200 MB
Total	< 55%	< 1.3 GB

The remaining ~45% CPU and ~14 GB of memory are available for DimOS Agent inference (local LLM or API calls) and future perception tasks (object detection, depth estimation, etc.), leaving ample headroom.

6 Future Roadmap

6.1 Platform Miniaturization — 5-inch Configuration

The core value of autonomous drones lies in high agility and deployment flexibility. Typical target scenarios include dense-forest search and rescue, tunnel/mine inspection, indoor structural survey, and military reconnaissance.

The current 10-inch platform provides large payload margin and low iteration cost as a validation testbed, but its footprint is not optimal for the scenarios above. After validating algorithm feasibility, it is recommended to miniaturize the platform to a 5-inch configuration for better fit with real-world deployment needs.

10-inch vs 5-inch Comparison

Dimension	10-inch (Current)	5-inch (Planned)
Form Factor	~450 mm wheelbase	~220 mm wheelbase
Maneuverability	Moderate	High
Gap-crossing Ability	Limited indoors	Excellent (fits through narrow spaces)
Payload Margin	Large (easy sensor expansion)	Medium (hardware must be lean)
Platform Cost	Higher	Lower
Primary Use	Algorithm validation, feature integration	Field deployment, real-world missions

6.2 Hardware Simplification — Integrated FC Solution

Domestic manufacturers now offer all-in-one flight controller and ESC solutions designed specifically for autonomous drones, significantly simplifying the hardware topology compared to discrete components:

• Integrated FC: Combines flight controller, RC receiver, and telemetry module in one unit, reducing wiring and external modules.
• Integrated ESC: Built-in 5V/12V BEC output to directly power the onboard computer and sensors without a separate BEC.

Hardware Topology Comparison:

Current 10-inch (Discrete)          Planned 5-inch (Integrated)
──────────────────────────          ───────────────────────────
Flight Controller (Pixhawk)         Integrated FC (FC + RC + Telemetry)
  + RC/Telemetry combo module       Integrated 4-in-1 ESC (with BEC)
  + 4-in-1 PDB                             Motors × 4
  + 4× Individual ESCs
  + Motors × 4

Eliminating 3–4 separate modules reduces total weight by approximately 80–120 g and improves reliability by reducing connector count.

6.3 Battery Upgrade — Semi-Solid-State Cells

Conventional LiPo batteries offer ~200–250 Wh/kg energy density. Semi-solid-state lithium batteries deliver significantly higher energy density while improving safety:

Battery Type	Energy Density	Notes
Conventional LiPo	200–250 Wh/kg	Current standard, low cost
Semi-solid-state (commercial)	300–330 Wh/kg	Mass-produced by select vendors, higher cost
Semi-solid-state (custom)	500 Wh/kg	Requires vendor customization, highest cost
Current (EVE JP40)	204 Wh/kg	High-performance 21700 cylindrical cell

The planned 5-inch platform uses commercial semi-solid-state cells at an assumed 310 Wh/kg energy density.

6.4 5-inch Platform Performance Estimate

Weight Breakdown

Component	Weight
Airframe (frame, integrated FC, integrated ESC, motors)	~200 g
Jetson Orin Nano (replacing Orin NX)	~150 g
Livox Mid-360	300 g
Miscellaneous (mounting hardware, wiring, damping)	~100 g
Airframe Total	~750 g
Semi-solid-state battery (310 Wh/kg, 428 g)	428 g
Total AUW	1178 g

Battery Parameters

Parameter	Value
Capacity	5600 mAh
Energy Density	310 Wh/kg
Battery Weight	428 g
Total Energy	132.7 Wh

Hover Endurance Calculation

• Required hover thrust: 1178 g
• Single motor hover load: 1178 ÷ 4 ≈ 295 g
• 5-inch motor thrust efficiency: 4.11 g/W
• Hover power (motors): 1178 ÷ 4.11 ≈ 286.6 W
• Onboard electronics (Jetson Orin Nano + LiDAR + misc.): 50 W
• Combined total power: 286.6 + 50 = 336.6 W
• Theoretical hover time: 132.7 Wh ÷ 336.6 W × 60 ≈ 23.6 min
• Corrected theoretical endurance (× 0.9): ≈ 21.3 min

Comparison with 10-inch Platform

Metric	10-inch (Current)	5-inch (Planned)	Change
Total Weight	2070 g	1178 g	−43%
Battery Energy	114.2 Wh	132.7 Wh	+16%
Motor Hover Power	300.4 W	286.6 W	−5%
Onboard Electronics	50 W	50 W	—
Combined Total Power	350.4 W	336.6 W	−4%
Theoretical Endurance	19.6 min	23.6 min	+20%
Corrected Endurance	≈ 17.6 min	≈ 21.3 min	+21%

Although the 5-inch motor efficiency (4.11 g/W) is lower than the 10-inch (6.89 g/W), the 43% weight reduction and 16% higher battery energy yield an overall endurance improvement of ~21%.

6.5 Roadmap

Phase 1 (Current)              Phase 2
10-inch Validation  ──────►  5-inch + Semi-solid Battery + Integrated FC
  PX4 + ROS 2                  Total AUW ≤ 1200 g
  EGO-Planner                  Corrected endurance ≥ 21 min
  DimOS Agent                  Wheelbase ≤ 220 mm
  Indoor validation            Real-world mission deployment

---

Synced from DIM-1088

[dimos-bot]

weyne (via Linear):
2026.06.30

Done

• Completed the initial design draft for the autonomous drone platform.

To Do

1. Communicate with the DIMOS team to finalize the autonomous drone solution.
2. Complete hardware procurement (or arrange to borrow the required hardware).
   ‌‌‌

[dimos-bot]

jetson (via Linear):
autonomous_drone_system_plan.md

@ivan hey please check proposal by @wei.jiang
‌‌‌

[leshy]

Above everything looks generally ok, I validated your flight time estimates, a few notes on above.

small corrections

1. we will not use ros for this, we have our own mapping framework that will take lio output.

2. Telemetry system proposed does not support video downlink - I believe this is a small mistake. ideally we want bidirectional communication and video downlink to the computer, so we can run visual algos. Something like SYIY should cover this. For development we might use direct jetson wifi connection to stream pointlio/fastlio output over zenoh - so should make sure it has a pcie wifi card installed.

longer flight time / heavy payload frame option

Alternatively, for more options with development, outdoors testing, I'm sharing my long flight time design (50 minutes, 1kg capacity) BOM and you can decide which build to go for initially.

Propulsion + Frame + Power

Item	Product	Unit ¥	Qty	Line ¥	~$
Frame	Custom H-ladder CF tube build (399 g) or Tarot IM650 ¥559	~512	1	512	72
Motor	T-Motor MN5006 KV300 Antigravity	~460	4	1,840	257
ESC	T-Motor ALPHA 40A 6S FOC	~385	4	1,540	215
Prop	T-Motor NS16×6.1 or 5.4 CF (pair) — 2 build + 2 spare	~340	4 pr	1,360	190
Cells	Molicel P50B 21700 (6S4P = 24)	42	24	1,008	141
BEC	Matek BEC12S-PRO (9–55 V → 5 V)	~150	1	150	21
Power/wiring	6S PDB + XT60/XT90 + 12 AWG silicone	~150	1	150	21

Avionics + Payload

Item	Product	Unit ¥	Qty	Line ¥	~$
FC	Holybro Pixhawk 6C (浩翼; base, no PM)	1,176	1	1,176	164
GPS	Holybro M10 GPS / compass	~330	1	330	46
Compute	NVIDIA Jetson Orin Nano Super Dev Kit (8GB)	1,999	1	1,999	280
Datalink	SIYI HM30 (air+ground, FHD video+telem)	~1,799	1	1,799	252

I would also consider RTK for this

[dimos-bot]

jetson (via Linear):
From @ivan - somehow his comment didn't get synced from Github
‌‌‌

[dimos-bot]

weyne (via Linear):
I'm guessing you're looking to build a drone geared toward industrial applications, rather than something like an indoor lab-grade micro drone. So here are my suggestions on hardware selection:

1. Motors — For safety reasons, waterproofing and dust protection are critical for outdoor operations. We could consider motors with these features, such as the SIYI E3 Propulsion Kit (https://siyi.biz/zh/product/all-in-one-propulsion-kit/e3/). Its efficiency performance is comparable to the Tmotor MN5006, but it offers modern CAN-based throttle control, historical data logging, and comes as a complete all-in-one solution.

2. Airframe — A generic foldable frame would work well: 15-inch props, wheelbase around 700mm, with folding propellers. This makes transport and field testing much more convenient.

3. Battery — I haven't personally used the Molicel P50B 21700 cells. At ¥42 per cell, they're quite expensive, but we could purchase a batch to test whether they're truly worth the price. Currently on the Xianyu (second-hand) app you can buy EVE 50PL cells — these are tested and proven to be reliable 21700 5000mAh batteries.

4. Power System — We might consider CUAV's PDB (https://doc.cuav.net/power-module/can-pdb/en/). Paired with a CUAV flight controller, a single board provides a complete flight control and power solution. Right now, using a Pixhawk 6C + PM module + Matek DC-DC converter tends to make the hardware wiring messy. Additionally, CUAV's flight controllers comply with the Pixhawk hardware standard. If you later want to implement pure inertial return-to-home as a safety strategy, the flight controller can be upgraded to the CUAV X7+ Pro, whose primary IMU is the ADIS16470 — even without any other positioning data sources, inertial navigation alone delivers decent performance.

5. Compute Platform — The Orin Nano is certainly a cost-effective compute platform, but if I recall correctly, it lacks a hardware video encoder. If the camera outputs raw data and you need to stream it, you'd be stuck with software encoding. So I'd still recommend the Orin NX 8GB.

6. Video/Data Link — The SIYI HM30 is an excellent solution for video transmission, and we can start the project with it. However, my original intent is to capture the video stream on the Orin NX and then push it over 4G along with telemetry — meaning 4G/5G would be the primary data link, with the traditional RC and telemetry radio serving as a safety backup. The ultimate goal is a fully cloud-controlled autonomous platform, which I feel aligns better with an agent-based autonomous execution model. After all, if the drone can truly perform missions autonomously, safely, and reliably, we won't need a pilot standing there with a radio watching it — or one pilot could manage multiple drones simultaneously.

As for the software architecture, the earlier design was mainly aimed at rapid validation. So the current roadmap would be: I could first deploy a SLAM + planner project on my own drone with a LiDAR (without DiMOS), test its usability and performance, and then migrate the planner into DiMOS. At that point, DiMOS can directly read LiDAR data, build maps internally, and work with the planner for navigation and obstacle avoidance. I'm also aware that DiMOS now has two navigation frameworks, but I'm not sure whether they're suitable for drones. So I still lean toward migrating battle-tested planner projects (analogous to nav frameworks) — for instance, the well-known EGO-Planner, ZJU's latest Diff-Planner, or HKU's Super-Planner.
‌‌‌

[dimos-bot]

Ivan (via Linear):
This all sounds very good to me, I offered my drone spec if helpful but I'm not insisting on my build and we are happy with any drone. Lidar/mapping/navigation is the important part here for us, the rest we are flexible on.

It's easy to build a new frame once we validate the build, so I'd let you completely own this and decide based on your preferences on which frame to build. large propellers I suggest are much more dangerous for initial tests as well

For next steps, please order the parts you need with @Jetson and split this work into separate tickets, like frame build, lidar integration etc, and we'll set a rough timeline.
‌‌‌

[dimos-bot]

weyne (via Linear):
I completely agree — at the early stage, we should focus as much as possible on feasibility. Assembling and debugging a drone takes a huge amount of time, and larger airframes are significantly harder and more dangerous to tune. So I think at this stage, we can first use my existing drone to: (1) complete planner project deployment and testing, (2) migrate the planner into DiMOS, and (3) test in-DiMOS navigation and obstacle avoidance.Once we've achieved the goals of the current phase, we can then consider building a drone that aligns more closely with industry application standards.

I'll follow up with Jetson later to finalize the project timeline and the procurement list.

‌‌‌

[dimos-bot]

weyne (via Linear):
2026-07-04 完成设备搭建、mid360接入、fast_lio部署、改装mid360航插接口。

‌‌‌

[dimos-bot]

weyne (via Linear):
Completed the deployment of the diff-planner project and the maiden flight test (which was not stable enough). Since the outer-loop control of diff-planner is on the ROS side, further PID tuning is required to stabilize the drone's flight. The flight debugging will be completed in the next two days.
‌‌‌

[dimos-bot]

Weyne (via Linear):
Currently, we have verified that the drone equipped with a MID360 LiDAR, using odometry from Faster‑LIO and feeding it to the PX4 flight controller, can perform position‑hold hovering indoors. The fixed‑point performance is good under normal conditions, but during aggressive maneuvers and rapid yaw changes, the response is somewhat slow. This week, we will focus on addressing these flight stability issues.

At the same time, we have determined that the immediate goal is to prioritize integrating the autonomous drone with DIMOS, so that we can see the effects more quickly. The planner’s path planning and obstacle avoidance functions are less urgent and will be addressed later.

飞书20260709-222637.mp4

飞书20260709-222642.mp4
‌‌‌

[dimos-bot]

Weyne (via Linear):
Over the weekend, I tested the integration of MID360 with DimOS and compared PointLIO and FastLIO2 on an Orin NX. I kept the output configurations as consistent as possible in order to compare their resource consumption. The results are provided in the appendix. Since there is no significant gap between the two, and PointLIO can output native twist data while the current Orin NX 8G can handle the load, I will proceed with PointLIO for now.

Current Completed

- MID360 integration, video streaming, and remote control in DimOS. (Flight testing is still pending.)

Plan for This Week

- Complete functional testing of the DimOS drone integration.

- If the functional tests pass, migrate the temporary TUI panel capabilities into dimos-viewer and design more general-purpose control buttons for commands beyond drone velocity control.

POINTLIO_FASTLIO2_NX_BENCHMARK_REPORT.md
‌‌‌

[dimos-bot]

Ivan (via Linear):
this is great, as you go into lio/mapping stuff you can check out our nav tooling

https://github.com/dimensionalOS/dimos/wiki/nav-tooling

specifically fastlip/pointlio tuning (you can record a pcap and easily replay it to see the map)

as well as global map functions
‌‌‌

[dimos-bot]

Ivan (via Linear):

‌‌‌