- DARPA Robotics Challenge
Robots that are capable of serving as first responders will allow for faster, safer, and more capable service in the aftermath of disasters. The DARPA Robotics Challenge (DRC) pushed robots to leave the security of the lab, taking on new roles in challenging, unpredictable disaster response environments. As many disaster scenarios take place in manmade environments, bipedal humanoid robots provide a means of navigating such environments while using tools originally designed for human responders (Kajita, Hirukawa, Harada, & Yokoi, 2014). Although traditional wheeled platforms exhibit advantages over bipeds, such as inherent stability, high payload capacity, and ease of con- trol and state estimation, bipeds offer a unique potential for mobility and flexibility in a variety of environments. The DRC effort focused on developing the key enabling technologies to field a robot capable of robust bipedal locomotion and manipulation in unstructured environments, while supporting varying levels of autonomous operation.
In addition to improved locomotion, ESCHER incorporates a new software system. Previous challenges in developing and integrating higher-level software on THOR led to using the robot operating system (ROS) to leverage open-source software and integrate with the existing whole-body control framework. This allowed Team VALOR’s software team to collaborate with Team ViGIR for perception, human-robot interaction, and path and manipulation planning. This use of open-source packages enabled Team VALOR to deploy unique research contributions while rapidly building up key subsystems. When incorporated onto ESCHER, a robot capable of both compliant locomotion and semi-autonomous behaviors was realized, resulting in a platform capable of walking and manipulation robust enough for the DRC Finals. ESCHER is one of only four custom bipedal humanoids designed for the DRC Finals.
video: Introducing ESCHER
2. Whole body Control
Torque-controlled bipeds and quadrupeds are becoming increasingly prevalent as researchers attempt to mimic the speed and adaptability of locomotion behaviors found in nature. This has created an interest in compliant locomotion strategies and inverse dynamics approaches that are robust to the unexpected forces and unmodeled dynamics encountered by hardware systems operating in real environments. The problem is further complicated in the field of humanoid robotics, where whole-body control is required to implement multi-objective behaviors such as maintaining balance while aiming a fire hose or operating a drill.
Implementation of optimization based whole-body control and compliant locomotion on a new torque-controlled humanoid is used. We propose an efficient quadratic program formulation to solve the inverse dynamics problem given frictional contact constraints and joint position / torque limits. The centroidal dynamics are stabilized using a novel momentum controller based on the time-varying DCM dynamics, implemented for the first time on hardware. These components are assembled into a general framework for compliant locomotion on uneven terrain. Successful whole-body control of the humanoid platform is achieved using a “simple” joint impedance controller that combines high performance torque control with low-gain velocity feedback. To improve the stability of the presented inverse dynamics-based approach, joint velocity set points are computed from the optimized joint accelerations and tracked using pre-transmission velocity estimates.
3. Multi Contact Control
One of the primary advantages of of bipedal platforms is their ability to traverse complex terrains, and the natural extension of this, humanoids with two dedicated manipulators, allows for vastly impressive feats of dexterity. However, in order to perform many of the tasks that humans so thoughtlessly complete daily, it is important that humanoid robots be able to interact with their surroundings beyond a purely locomotory “one foot in front of the other” approach. Though many independent systems will need to be integrated to begin to match the proprioceptive acuity of human beings, proper adaptive strategies for integrating multi-contact scenarios into the robot’s controller are necessary to expand the locomotory capabilities of the platform while maintaining the distinct advantages of having versatile limbs not explicitly dedicated to locomotion.
When extrapolated to the three-dimensional case, many controls procedures can become significantly more complicated. Indeed, since many hand contacts tend to be more for stability than support, only small forces are required at those contact points. Humans use this advantageous force distribution to their advantage, frequently grabbing rails, handles, or merely placing their hands on flat surfaces and using a combination of normal and tangential forces to generate appropriate stability to the desired locomotory path. However, humans are able to easily detect and mitigate slips of the hands, utilizing the natural tactile abilities of the hands to prevent undesired slippage that could lead to a dangerous fall.
Though there are some existing implementations of tactile sensors used to detect slip, the cognitive ability to measure and accommodate slippery surfaces during operation has been undervalued in humanoid robotics. Especially in non-gravity normal contact scenarios, slip is possible and perhaps even likely if friction considerations are not explicitly made in the whole body controller. Though it is generally sufficient to utilize an extremely conservative estimate of the friction coefficient, this does severely limit the potential applications of force possible at the contact point. It can be difficult to provide accurate information regarding friction before contact due to the relatively small surface properties that most profoundly affect the resulting surface forces, especially in more complex environments where surfaces vary widely and visual systems may not be able to accurately detect minor characteristics. For this reason, it is useful to accurately approximate the friction coefficient of the contact surface before or while loading the contact end effector, and providing compensatory methods in the event of slippage.