Center for Robotics, Sentry Micro Robot

Micro-Robot Sentry: Kshudra

Automated surveillance applications widely use cameras mounted at inconspicuous sites with pan/tilt/zoom capabilities. Many security critical applications however, span larger spaces where a more mobile camera application is needed. Furthermore, often the presence of sounds in the remote location can be significant, and may key in the attention of the humans monitoring the camera banks.

Another critical need is that of partially automated identification of points of attentive focus in such systems. Thus, the system can optionally learn a "security model" based on which it can identify suspicious events and attract the attention of the human monitor, activate alarms, move in a direction for best tracking the intrusion, and also signal other devices, e.g. to initiate video recording.

These are the two needs addressed by our micro-robot Sentry Kshudra (3"x3"x4")which is designed to be inconspicuous and to pack a powerful set of sensory modalities including vision and sound, and also the capability of processing its inputs to identify areas of possible interest. The origins of this robot are in the micro-robots used in the international soccer robot competition MIROSOT.

This idea was originally developed as a project by Tamhant Jain and Nishant Agrawal.

Top

Design

Kshudra follows a modular approach since it facilitates an uncoupled development of the different subsystems, and lends greater versatility to the functions that can be mounted on this platform. The block diagram below illustrates the various modules.

Particularly significant in this design is the option of off-board vision processing which enables intelligent detection of special events such as sounds, irregular movements (esp. humans in motion) and other needs specific to the particular security environment. This system is mounted on the host system where the images are being received and may also be coupled with other cameras and monitoring devices. This setup offers two different imaging systems with different capabilities (based on the Matrox Meteor or the Cognachrome systems). The system can track objects by colour, shape, and by motion, and return the positions and velocities of these objects which are then processed using environment-specific security models.

With or without this off-board vision model, Kshudra can make decisions about where to move in its given space. It uses a priori information about the environment given in the form of a 2-D map, as well as immediate image-based obstacle avoidance based on optical flow, so that temporary obstacles such as boxes and crates etc can also be avoided. The system uses completely wireless communication based on the double-frequency R/F communication using RS232C protocol. We are using Radiometrix low power transceivers, which are specifically made for applications of this scale. The transceivers work up to the range of 100m, and can be re-transmitted further using relay stations. These are compact, reliable and have low emissions (FCC compliant).

The robot itself is about 7.5-cm cube in size. The main microcontroller (Intel 8051) receives commands from the host computer through the serial port and after interpreting them, controls the robot directly using a motion controller implemented on the micro-controller itself. Robot locomotion is achieved using two motors through differential drive. The system is able to achieve high speeds of upto 1.5m/s and is capable of overtaking a walking human.

mainboard.jpg (18353 bytes)

Onboard circuitry for micro-robot Kshudra. 8081 family ATMEL microprocessor - right-center; Radio transceiver chip - bottom right. (matchstick = 5cm).

p An earlier version of this robot called Vocs_Miro won the second-best design award at the ELECRAMA-99 conference in Bombay. This robot also employed a speech interface that allowed it to respond to commands received via a telephone. This option requires the installation of a commercial voice recognition system (Via Voice) and training of the individual controlling voices.

Commercial Aspects: Kshudra is being developed in a basic package which is expected to cost about Rs. 25,000 and will include basic image and sound transmission and tracking capabilities. The advanced version with automatic alarm and security-model based visual intrusion detection will cost approximately Rs. 70,000 per robot including the off-board image processing system. About 40% of these projected costs are development costs at a stable volume of 1000 pieces p.a.

Competition: While other sentry robots (e.g. Savannah from Real World Interfaces) are available, these are typically far larger and therefore subject to being detected and avoided. Also they cost about $5,000 which is about five times the projected cost of Kshudra.

Volumes and Projected Turnover: The security market is estimated at $20 billion worldwide and growing at 20-50% annually worldwide, and nearly doubling annually in India, especially in cities like Delhi or Bombay. This includes of course many other items, but it would be reasonable, given the cost advantage of such a system, that volumes supported would be about 200 units the first year (mostly in defence, nuclear and other sensitive sites in India), 1000 in the second (India and worldwide), with a growth rate of nearly doubling for the first four years, to reach a stable volume of about 10,000 per annum at an average price of $2,000, so as to reach a turnover of $20 million in five years.

Top