Sree Prasanna Rajagopal

My research motivation is building bio-inspired cobots. Particularly, I’m interested in motion and manipulation design with soft actuators for semi-autonomous, human-assisted robots.

[[resume]]

June 2015 to April 2019

Avenger30 - a gunner assist system

worked with:

Tonbo Imaging

page.imgalt

Summary

Avenger30 was designed and developed in Tonbo Imaging, Bangalore. Tonbo Imaging designs, builds and deploys advanced imaging and sensor systems to sense, understand and control complex environments. Our team was a mixed and versatile collection of mechanical, electronics, computer vision, and design engineers, with the core focus on developing a Dual axis FOG stabilized electro-optic sight with cooled MW IR thermal imager, low light CCD camera, laser range finder for surveillance, reconnaissance and target acquisition. It provides stabilised visual feedback via a housed camera while actively rejecting external motion disturbances.

Details

avg30_mounted.png

This crucial system enables the human operator to defend and neutralize threats from behind the safety of armour. Such a forward deployed system required immense dependability on both hardware and software aspects. Particularly I worked on the fault tolerance, robustness, and the decentralization of the control systems. I deployed motion control algorithms such as sensor fusion techniques, gyro bias correction methods, and cascaded motion controllers for line-of-sight stabilization.

Milestones

avg30_flowchart.png

Literature review:

Competitor gimbals/PTUs specifications, possible drive architectures, aesthetics.

Specifications:

We aimed for:

Size estimation:

To select appropriate motors, we first needed a reliable inertia estimate of the system, accounting for the eventual extra weight of balancing weights.

Sensor selection:

Broad criteria for selecting sensors was ease of integration and acquisition rates. Higher acquisition rates meant that control loops could be pushed to higher rates later if required.

Platform selection:

The operating torque is a consequence of estimated inertia and required acceleration. The computing platform should have supporting motor control circuits capable of high PWM frequencies, amperage requirements, available interfaces for the selected sensors. Higher the PWM frequency, the more freedom to select our motors, since the L/R ratio of the motor sets a lower bound on the required PWM frequency.

Firmware architecture:

My focus for firmware development was on modularity and fast iteration. The ultimate aim was to create lego blocks of control modules that could be mixed and matched quickly and reliably.

Component selection verification:

Are these the right encoders, gyros, motors for the chosen specifications? We designed tests to estimate maximum possible control loops frequency. We estimated maximum achievable acceleration of the system, and the response bandwidth possible with the chosen motors. We created test rigs to test our specifications.

Control engineering:

After components freeze, we pushed the system against the chosen specifications, particularly the stabilization accuracy. This phase was an iteration of control loops, tuning, and modeling work.

Auxiliary debug tools:

JTAGS are messy. They restrict development speed. At this point, I developed tools for faster control loop testing. The end result was a JAVA debugger that could interface with the AVENGER through a serial port/ethernet port and expose internal control variables in real time on a moving graph. It also provides access to the control loop parameters. This allows engineers to quickly make changes to the control loops without having to flash code every time.

Final performance:

We achieve 30 micro radians of stabilization accuracy. That’s less than 5 pixels of variation at a 2000 m distance.

tags: defense - gunner