The Stanford Testbed of Autonomous Rotorcraft for Multi-Agent Control (STARMAC) is a multi vehicle test bed used to demonstrate new concepts in multi agent control on a real-world platform. In order to make such a testbed easy to use, we focused on a small and light, lowcost design, which presented numerous opportunities for innovative work. STARMAC consists of up to eight quadrotor vehicles that autonomously track a given waypoint trajectory, and the main goals of the project lie in two phases:
- Phase 1: Initial Vehicle and Testbed Design for Proof-of-Concept Flights
- Phase 2: Complete Vehicle and Testbed Redesign for full functioning testbed
- Phase 3: Multi Agent Control Demonstration
Starmac I Vehicle overview
The base vehicle was originally an off-the-shelf four-rotor helicopter known as the DraganFlyer IV, which has a total of 1.0 kg of thrust and can sustain hover for about ten minutes at full throttle ( how quadrotors work ). The open-loop vehicle dynamics are unstable and extremely fast, making the X4 flyer almost impossible for humans to fly. An existing onboard controller slows down the system dynamics to about 5 Hz and adds damping, making it pilotable by humans.
State sensing was performed using 3-axis gyro, accelerometer and magnetometer information, as well as sonic ranging for altitude and differential GPS for position. In order to make the vehicle capable of achieving the goals of the STARMAC project, a complete redesign of both the frame and the onboard electronics was performed. The frame stiffness was greatly increased to provide a reduced vibration environment and to increase crash survivability. The onboard electronics were replaced with our own custom printed circuit board (PCB) so as to have complete control over motor commands, power supply and sensor measurements.
In the fall of 2004, we successfully performed autonomous hover of two STARMAC I quadrotors outdoors (see video here). However, 4 significant limitations were observed, and as such, a redesign of the base vehicle was required.
Starmac II Vehicle overview
The four major modifications for Starmac II are outlined here.
1. Thrust Capabilities:
Brushless motors and more rigid plastic propellers were combined to double the efficiency and increase the total available thrust fourfold, resulting in a total of 4 kg of thrust for each vehicle. STARMAC I was close to its payload limit with only 1 kg of thrust, operating at 75% full throttle at hover. The improvements to vehicle thrust on STARMAC II enable the inclusion of larger batteries to further extend flight time, as well as the inclusion of additional sensors and computation resources.
2. On-board Computation Resources:
With the added lift capabilities, it was possible to significantly increase onboard computing power.A low level microcontroller board, called the Robostix, was added for inner loop vehicle control. For position estimation and control, the Stargate, an Intel PXA255 based single board computer (SBC) was added. Finally, a PC-104 computer was added running Windows XP for onboard processing of vision information, optimal path planning and other high level automation tasks.
3. Communication Reliability and Bandwidth:
The testbed communication channel was switched from Bluetooth to WiFi, enabling much greater bandwidth, moderate gains in range, and improved communication channel management through a wireless router at the basestation. The Stargate platform was pre-configured with a compact flash802.11bWiFi card, and field testing has revealed significant improvements in communication robustness between base station and vehicle as compared with the Bluetooth capabilities of STARMAC I.
4. Position Measurement Accuracy:
STARMAC II relied on fusing 1 Hz code phase differential GPS measurements with 76 Hz IMU data to estimate vehicle position, which resulted in accuracy no better than 3-5 m circular error precision (CEP). The resulting accuracy on position control could do no better than the position estimate, which represented hover location variability of five times the vehicle size. In order to keep vehicle weight and cost down yet significantly improve position estimate accuracy, it was decided to develop a 10 Hz carrier phase differential positioning system at Stanford, which resulted in errors of 2cm CEP using the low-cost Novatel SuperStar II platform.
Initial flight results for the STARMAC II platform are now available in the gallery, and an additional five vehicles are being built to expand the fleet to six vehicles. The significant improvement in capabilities outlined in this section will allow the STARMAC II vehicle to be used in real-world applications, with sufficient computational resources and excess lift capacity to tackle challenging tasks autonomously.