Introduction

The main focus of our research is in the exploration of the effectiveness of a group of small robots employing distributed sensing platforms. We believe that with proper coordination, a set of small, disparate entities can effectively achieve the functionality of a larger robot while retaining the ability to operate in unknown domains. However, with small size comes the disadvantages of limited mobility range, limited energy availability, and possibly reduced sensing, communication and computation ability due to size and power constraints. Because limitations in size are immediately extended to power and processing capabilities, it was realized early that our robots would have to coordinate and cooperate to achieve any useful tasks. To that end, we designed and constructed a team of heterogeneous, centimeter-scale robots and construction of a team of centimeter sized robots called Millibots. The robot team exploits modular sensing, processing and mobility to achieve a wide range of tasks that include mapping and exploration as well as support for fire rescue. From the operator’s point of view, the team can be controlled as a single logical entity. We focus on how the team can exploit communication and sensing to perform missions such as mapping, exploration, surveillance and support for rescue operations.

To achieve extended team operation in the face of uncertainty, we have developed a localization system that uses sonar-based distance measurements to determine the positions of all the robots in the group. In [1] we describe how the team coordinates sensing and action to utilize localization and maintain group position. Furthermore, by incorporating Bayesian techniques in data fusion, we are able to integrate individual robot sensing to generate composite area maps. This technique is further extended to merge ranging and heat information to detect and isolate potential warm bodies in a fire rescue scenario.

In this document we describe the most recent sensor platform for the Millibots: the Directional Infrared Detector Module (DIRM). Heat sources, such as open flames, hot zones or unconscious people, are usually worth to be explored. Particularly, detection of people and people activity could be a mission itself. Objects that generate heat also generate infrared radiation (IR), which can be easily detected. For this reason we designed a directional IR detector module (DIRM), which provides the Millibots with means of increasing their sphere of awareness when exploring an uncharted space (Figure 1).

The Directional Infrared Detector Module (DIRM)

The heart of the DIRM is a pyroelectric infrared sensor (often known as PIR sensor). Pyroelectric sensors are made of ferroelectric crystals that generate a surface electric charge when exposed to IR. However, a pyroelectric sensor only produces an electrical output when the level of incident radiation changes. Electrical output is a function of the rate of change in detector temperature rather than temperature itself. This characteristic is one of the reasons why pyroelectric sensors are widely used as thermal motion detectors for detecting people. However, if the incident radiation changes slowly, the electrical output of the sensor will be small and may not be detected regardless of the magnitude of the source. To enable the detection of immobile heat sources, we have designed a DIRM that is capable of detecting both mobile and immobile heat sources such as warm bodies or overheating equipment. The sensor is mounted to a rotary platform that smoothly rotates and sweeps the sensor across an arc of about 170 degrees. If the sensor points towards a heat source while sweeping the area, it will produce an electrical output in response to the change of incident radiation.

Design considerations

Although the concept of rotating pyroelectric sensors is simple, several factors complicate the detection process. First, pyroelectric sensors below a temperature of T_c, known as the Curie point, exhibit a large spontaneous electrical polarization. This behavior is analogous to piezoelectric materials and can cause the sensor to produce spurious signals when subject to acoustical or mechanical excitation. Smooth motor operation and proper detector isolation are required to minimize this effect. Secondly, electrically induced noise produced by the motor as it rotates affects the signal conditioning circuit of the sensor, consequently degrading the detection quality. We included a dedicated voltage regulator to filter out electrical noise and provide clean power to the module. For example, Figure 2 shows the current drawn from the servo while sweeping across a full 170 degrees arc. The current supplied by the battery increases considerably, from less than 10 mA to over 350 mA. This is a primary source of noise, which had to be taken into account. Finally, the third factor to consider is the incident radiation itself. We are primarily interested in the detection of heat sources. However, the pyroelectric sensor detects only a temperature change, regardless of whether this change was positive or negative. Although all bodies tend to reach a thermal equilibrium, under normal operation conditions it is likely to detect several temperature variations inside the same room. Consider for instance a normal office, in which the air conditioner is set to keep the room temperature at 25°C. The temperature inside the room on the average measures 25°C; however, we can measure warmer temperatures in areas closer to the computer monitor or a window bathed in sunlight, and cooler temperatures in areas closer to the vents. When the pyroelectric sensor is swept across this office, it produces a distinct output whenever the temperature of the zone it is crosses either cooler or warmer regions. For this reason, we adjusted the module so that it only produces a response from large variations in incident IR.

Fig. 2. R/C servo current

Functional description

Figure 3 shows a block diagram of the DIRM. A Fresnel lens captures the incident IR and focuses it towards the pyroelectric sensor increasing the sensitivity of the sensor and improving its directional response. The resultant signal passes through a low pass filter, which removes any high frequency noise generated by mechanical vibration. The output of the filter is then fed into a differentiator, which produces an output voltage proportional to the rate of change of the incident IR. The frequency response of this differentiator is also rolled off at high frequencies, further reducing the effects of undesired signals. The window comparator produces a logic output whenever the rate of change of incident IR exceeds a given set point. An 8-bit PIC16F84 microcontroller processes the logic signals and controls the rotating platform and reports information to the team leader.

Fig. 3. Block diagram of the DIRM

The DIRM can sweep an arc q =173° in approximately 5 sec. This arc is divided into 14 segments; each segment is associated with a zone z_i, , of length q /14 degrees. The microcontroller samples the output of the window comparator about 10 times per zone and records the corresponding digital output. Data is reported back to the team leader in a message containing two bytes, which contain the digital output associated with each one of the 14 bearing zones.

In order to further expand the sensing capabilities of the DIRM, we included a temperature sensor and a connector to which an infrared rangefinder can be attached.

The temperature sensor is built around a single-chip digital thermometer. It measures temperatures from –55°C to +125°C, with ±0.5°C accuracy from –10°C to +85°C. The microcontroller communicates with this sensor using a 1-Wire interface.

A Sharp™ GP2D02 infrared rangefinder [3] can be connected to the microcontroller as well. When this rangefinder is attached to the servo axis, along with the PIR detector, it is possible to extract range information to objects at several bearings. This sensor can measure distances to objects from 10cm to 80cm. The manufacturer claims this sensor to be impervious to color and reflectivity of reflective object. We have not characterized this sensor thoroughly yet. Moreover, further research needs to be done in order to fully exploit the combined capacity of both PIR and range sensors working together.

Fig. 4. Schematic diagram of the DIRM

Figure 4 shows the schematic diagram of the DIRM. The heart of the module is the PIC microcontroller. This device is in charge of parsing commands received from the host computer via the RF transceiver, controlling the servomotor, sampling the pyroelectric sensor output, controlling the temperature sensor and emitting signals for the localization system [2]. The microcontroller is In-Circuit Programmable; connector J2 is used for that purpose, along with the switches S1 and S2. The pyroelectric sensor (Eltec 442-3) is connected to J1. The signal conditioning circuit, a voltage reference and the window comparator are built around a single LM324 quad op-amp (U1). An inverter (U4) serves as a driver for the ultrasonic transducer (XTAL2). This transducer produces localization signals [2].

Heat source position estimation

The DIRM can only give a notion of the direction from which the heat sources originates, but not their range. To find the approximate position of the heat source, we need to combine several readings of the DIRM taken from different positions, or merge the bearing information with range data taken from other Millibots. As described in [1], we build occupancy maps by combining range information from different positions and time instances within an occupancy grid. By merging thermal information taken from several different locations, we can construct an equivalent thermal occupancy grid. Each thermal grid cell stores the likelihood that a heat source is present at that location. An occupancy value near 1 corresponds to a likely that position is occupied by a heat source. Similarly, a value near zero corresponds to the likelihood the area is free of a heat source. Since nothing is initially assumed about the environment, all the cells are initially assigned a value of 0.5 (equally likely to be contain or not contain a heat source). Following each measurement, the corresponding grid cells are adjusted using a Bayesian update rule based on a derived model of the sensor. A two-dimensional occupancy model generated for the DIRM in a single zone is shown in Figure 3. The sensor is modeled with Gaussian uncertainty in both range and angle. The profile shown corresponds to a DIRM positioned at the upper left and pointing to the lower right. The DIRM indicates a bearing in which a heat source is more likely to be found. With this information, and since the position and orientation of the DIRM are known, the team leader can send other Millibots equipped with different sensors to explore and pinpoint the location of the heat source. The next section describes the use of the pyroelectric detector in isolating and targeting victims in a fire scenario.

Fig. 5. Occupancy grid for the DIRM

Heat source localization

When working as a coordinated, mobile sensing platform, a team of Millibots can extract rich information in many new scenarios. Consider for example a team of firefighters working inside a burning building. While the firefighters may easily spots open flames, some areas may be difficult to reach or inspect. Moreover, reduced visibility makes it difficult to detect and assist potential victims. This scenario depicts a situation in which additional sensing provided by a team of roving robots can increase the effectiveness of the rescue effort. Figure 6 illustrates a team of Millibots exploring a space in the vicinity of a fallen victim. Several sonar robots have already explored the area and generated a map of the zone. The robots have passed a heat source but mapped it as another obstacle. Because of the poor angular resolution of the sonars, the corners are not clearly resolved in the map so the free corners look the same as the corner in which the victim is located.

Fig. 6. Heat source localization

Similarly, a thermal occupancy map would only indicate the direction of the heat source but provide no indication of the size or distance to the source. However, by fusing the information from both sensor types together, the team is able to assess that there is an object of interest located in the corner of the derived map. Furthermore, if properly equipped, the team can further exploit its heterogeneous nature by directing a robot with a video camera to return a snap shot of the area. By exploiting multiple sensor modalities, the sum of the team's combined experience can provides invaluable assistance in many areas.

References

[1] Grabowski, R., Navarro-Serment, L.E., Paredis, C.J.J. and Khosla, P.K., 2000. Heterogeneous Teams of Modular Robots for Mapping and Exploration. Autonomous Robots. (special issue on heterogeneous and distributed robotics). Volume 8, No. 3, June 2000. pp. 293–308.

[2] Navarro-Serment, L.E, Paredis, C.J.J., and Khosla, P.K. 1999. "A Beacon System for the Localization of Distributed Robotic Teams," In Proceedings of the International Conference on Field and Service Robotics, Pittsburgh, PA, August 29-31, 1999.

[3] http://www.sharpmeg.com/products/opto/pdf/gp2d02.pdf