In this paper, we study the value of using synthetically produced videos as training data for neural networks used for action categorization. Motivated by the fact that texture and background of a video play little to no significant roles in optical flow, we generated simplified textureless and background-less videos and utilized the synthetic data to train a Temporal Segment Network (TSN). The results demonstrated that augmenting TSN with simplified synthetic data improved the original network accuracy (68.5%), achieving 71.8% on HMDB-51 when adding 4,000 videos and 72.4% when adding 8,000 videos. Also, training using simplified synthetic videos alone on 25 classes of UCF-101 achieved 30.71% when trained on 2500 videos and 52.7% when trained on 5000 videos. Finally, results showed that when reducing the number of real videos of UCF-25 to 10% and combining them with synthetic videos, the accuracy drops to only 85.41%, compared to a drop to 77.4% when no synthetic data is added.

This paper proposes a novel technique to extract training data from free space in a scene using a stereo camera. The proposed technique exploits the projection of planes in the v-disparity image paired with Bayesian linear regression to reliably identify training image pixels belonging to free space in a scene. Unlike other methods in the literature, the algorithm does not require any prior training, has only one free parameter, and is shown to provide consistent results over a variety of terrains without the need for any manual tuning. The proposed method is compared to two other data extraction methods from the literature. Results of Support Vector classifiers using training data extracted by the proposed technique are superior in terms of quality and consistency of free space estimation. Furthermore, the computation time required by the proposed technique is shown to be smaller and more consistent than that of other training data extraction methods.

This paper proposes a novel technique for segmenting the ground plane and at the same time estimating the occupancy probability of each point in a scene. Using a stereo camera rig, our system first calculates a disparity map and transforms it to a v-disparity map, which is then filtered and processed to generate a corresponding probability field. The probability field generated is then used for precise segmentation of ground planes as well as for the generation of occupancy grids. Unlike what is proposed in the prior art, our system requires minimal initialization and is independent of the stereo sensor characteristics as well as the parameters of the disparity algorithm. More importantly, our technique does not require any prior assumption about the terrain visual characteristics. Experimental results using sequences of images from two different data sets are presented to validate the proposed methods.

This paper is concerned with gait generation for the simplest underactuated multi-link robot. This robot is realized by a two-link planar system that locomotes in a viscous environment due to the existence of friction pads along its body. The proposed solution of the motion planning problem yields both a control gait for the inter-link angle as well as the location of the friction pad to produce a desired motion of the system.

In this paper, we take a rather new approach to solving the motion planning problem for principally kinematic systems, that is, systems whose dynamics is completely specified by the right number of nonholonomic constraints. Given a planar curve in the Special Euclidean group, we analytically solve for the base variables that locomote the principally kinematic system along the specified curve.

Small robots have the potential to access confined spaces where humans cannot go. However, the mobility of wheeled and tracked systems is severely limited in cluttered environments. Snake robots using biologically inspired gaits for locomotion can provide better access in many situations, but are slow and can easily snag. This paper introduces an alternative approach to snake robot locomotion, in which the entire surface of the robot provides continuous propulsive force to significantly improve speed and mobility in many environments.

We present a refined, second-generation design, construction and integration, of a compact hyper-redundant snakelike robot, called "Woodstock." This robot has substantial advantages over our previous design iteration, "Snoopy," in terms of cost and performance. The robot is composed of six actuated universal joints which are serially chained to construct a twelve degrees of freedom snake-like robot optimized for strength and compactness. Any joint in the robot is strong enough to produce a torque that is capable of cantilevering the entire robot. This paper also presents the low-level system- control architecture, which is based on a high-speed RS-485 data bus; this allows the entire system to be operated with only two power and two data wires. The system is controlled from a remote computer on a wireless network and can also run over the Internet.

In this paper we generate gaits for dynamics systems that are subject to non-holonomic velocity constraints. These systems are referred to as mixed non-holonomic systems. The motion of such systems is governed by both the non-holonomic constraints acting on the system and a system of differential equations constraining the evolution of generalized momentum. We propose a method that utilizes both governing motions, that is, satisfying all the constraints and instantaneously conserving momentum along un-restricted directions, to generate gaits for systems like the snakeboard, which belongs to the family of mixed non-holonomic systems. We accomplish this by defining a new scaled momentum variable. This scaled momentum allows us to easily explore the design of gaits that causes momentum to evolve such that a desired non-trivial motion results

This paper investigates how to generate cyclic gaits for multi-bodied isolated mechanical systems whose configuration space is represented by a trivial fiber bundle. We describe how to generate gaits in the base space of the fiber bundle, or the shape space of the robot on which we assume full control. Such gaits are guaranteed to generate a non-zero motion along the fiber space, i.e., a net change in the position of the robot, while making sure that the robot’s shape is unchanged after a complete cycle. The gait generation technique presented in this paper is intuitive; it involves dividing the base space into well defined regions and devising a set of simple rules on how to generate curves in such regions. Not only do such curves guarantee non-zero position change but also do allow for gait optimization.

This paper presents a novel compact design for a two degrees of freedom (DOF) joint mechanism. The joint is optimized for compactness, strength and range of motion which makes it ideal for constructing spatial or three-dimensional hyper redundant robots. We also identify and classify various prior joint designs that led to the development of this new concept. Finally, we present the joint forward kinematics, and force and torque calculations to verify the joint's range of motion and mechanical advantage.