Research ArticleCOLLECTIVE BEHAVIOR

Minimal navigation solution for a swarm of tiny flying robots to explore an unknown environment

See allHide authors and affiliations

Science Robotics  23 Oct 2019:
Vol. 4, Issue 35, eaaw9710
DOI: 10.1126/scirobotics.aaw9710
  • Fig. 1 Hardware specifications and comparison.

    (A) Crazyflie 2.0 with the flow and multi-ranger expansion decks and (B) the autopilot (STM32F4) compared with the specifications of the NVIDIA TX2, the Odroid-C2, and a laptop (Dell Latitude E7450). Note that the Dell specifications do not fit within the chart (as indicated with the top triangles).

  • Fig. 2 Main concept of the SGBA.

    (A) A simplified state machine of SGBA derived from the one presented in Materials and Methods. (B) Outbound travel of SGBA. The purple shading illustrates the local signal strength around each drone used for intra-swarm avoidance. (C) Inbound travel. The pink shading represents the signal strength of the wireless beacon at the ground station to which the drones navigate. The interswarm avoidance is still active on the inbound flight but is not depicted. The fuchsia arrow at each drone’s position illustrates the robot’s estimated direction to the beacon.

  • Fig. 3 Simulation results.

    The results of the simulation environments with (A) a representation of the ARGoS simulator and the modified simulated foot-bot. Two example environments and trajectories are shown for (B) six robots and (C) for four robots. (D and E) The results of 2, 4, 6, 8, or 10 robots in 100 procedurally generated environments for each configuration, in the coverage (not including nonaccessible areas), and the return rate. Three types of coverage are shown in (D): coverage total (area covered by all robots), coverage returned (area covered only by the robots that have returned), and coverage per robot (area that a single robot has covered). The exact computation of the covered area can be found in text S4.1. Last, in (E), the return rate is shown, i.e., the portion of robots that successfully returns to the base station after exploration. Both bar graphs of (D) and (E) show the mean as the SD, of which the specific values can be found in text S4.1.

  • Fig. 4 Real-world results.

    The results of the real-world experiments with (A) a representation of the environment used and the Crazyflie 2.0’s with the necessary expansion decks. Several example trajectories are shown for (B) four robots and (C) for six robots from their onboard odometry (adjusted by means of the external cameras). The results of two, four, or six robots for five flights in each configuration are shown in (D) for the coverage (not including the nonaccessible areas in gray) and in (E) for the return rate (cyan bars), with a pie chart additionally indicating the percentages of real-world–related issues, which prevented a successful return to the base station. Both bar graphs of (D) and (E) show the mean as the SD, of which the specific values can be found in text S4.2.

  • Fig. 5 Proof-of-concept search-and-rescue mission.

    The results of the experiment in which the Crazyflies carry a camera to detect victims in the environment. (A and B) Screenshots of the external cameras capturing the Crazyflies during their flight. (C) Trajectory of the four Crazyflies (inferred from the onboard and external camera). (D and E) Screenshots of the onboard Hubsan camera, with the two human-shaped silhouettes captured during the exploration flight.

  • Fig. 6 SGBA FSM.

    (A) FSM of the SGBA with (B) a legend of symbols. Its individual subsystems are illustrated as follows: (C) Outbound navigation in which the robot attempts to follow its goal heading, following obstacle contours on the way, and (D) local direction preferences based on angle of attack, i.e., the angle that the robot’s trajectory makes with the wall and the principle of the loop detection. The addition to the state machine of the gradient search toward the beacon at the home location for the inbound travel is given in blue, with (E) the gradient search method during the straight parts of the wall following. Here, the robot tries to estimate the direction toward the beacon by integrating information on the received signal strength based on its heading along the way. Swarm coordination addition to the state machine for the outbound flight (green), where (F) shows that the robot will change its goal heading if another drone (with higher priority) has its preferred heading in the same direction. In case the drones are even closer, the one with the lowest priority will move out of the way completely for both inbound and outbound travel.

  • Fig. 7 Hardware and communication specifics.

    (A) Crazyflie used for outbound and inbound travel and (B) the assembly used for the video recording of the environment. (C) Components on the Crazyflie, including weight and battery consumption. (D) Total of six Crazyflies used including six Crazyradio PAs and (E) the communication scheme shown for the six-drone experiment. Here, a counter is regulating when the drone will transmit a message (msg) to another drone (for counter 1: drone 1 to 2, drone 2 to 3, etc.). Between the regulated counter, the drone transmits its message to another drone with a time offset based on its ID. Six PAs were used for the six communication channels to receive logging of the Crazyflies for statistics; however, these can be replaced by one if no telemetry is required.

Supplementary Materials

  • robotics.sciencemag.org/cgi/content/full/4/35/eaaw9710/DC1

    Text S1. Real-world test environment

    Text S2. RSSI measurements

    Text S3. From odometry to trajectory

    Text S4. Analysis and statistics

    Text S5. SGBA implementation details simulation versus real world

    Text S6. SGBA submodule analysis

    Fig. S1. Overview of the real-world environment.

    Fig. S2. RSSI measurements.

    Fig. S3. Odometry versus trajectory.

    Fig. S4. Transcripts hallway video.

    Fig. S5. Odometry correction.

    Fig. S6. Coverage calculation simulation.

    Fig. S7. SGBA simulation versus real world.

    Fig. S8. Simulated collisions SGBA.

    Fig. S9. Loop detection check.

    Table S1. Statistics of the simulation tests.

    Table S2. End status of the real-world tests with two drones.

    Table S3. End status of the real-world tests with four drones.

    Table S4. End status of the real-world tests with six drones.

    Table S5. Coverage of the real-world tests with two drones.

    Table S6. Coverage of the real-world tests with four drones.

    Table S7. Coverage of the real-world tests with six drones.

    Table S8. Statistics of the real-world tests.

    Table S9. Real-world collisions.

    Movie S1. Video six-drone test configuration.

    Movie S2. Video four-drone victim search.

  • Supplementary Materials

    The PDF file includes:

    • Text S1. Real-world test environment
    • Text S2. RSSI measurements
    • Text S3. From odometry to trajectory
    • Text S4. Analysis and statistics
    • Text S5. SGBA implementation details simulation versus real world
    • Text S6. SGBA submodule analysis
    • Fig. S1. Overview of the real-world environment.
    • Fig. S2. RSSI measurements.
    • Fig. S3. Odometry versus trajectory.
    • Fig. S4. Transcripts hallway video.
    • Fig. S5. Odometry correction.
    • Fig. S6. Coverage calculation simulation.
    • Fig. S7. SGBA simulation versus real world.
    • Fig. S8. Simulated collisions SGBA.
    • Fig. S9. Loop detection check.
    • Table S1. Statistics of the simulation tests.
    • Table S2. End status of the real-world tests with two drones.
    • Table S3. End status of the real-world tests with four drones.
    • Table S4. End status of the real-world tests with six drones.
    • Table S5. Coverage of the real-world tests with two drones.
    • Table S6. Coverage of the real-world tests with four drones.
    • Table S7. Coverage of the real-world tests with six drones.
    • Table S8. Statistics of the real-world tests.
    • Table S9. Real-world collisions.
    • Legends for movies S1 and S2

    Download PDF

    Other Supplementary Material for this manuscript includes the following:

    • Movie S1 (.mp4 format). Video six-drone test configuration.
    • Movie S2 (.mp4 format). Video four-drone victim search.

    Files in this Data Supplement:

Stay Connected to Science Robotics

Navigate This Article